This article analyzes David Sacks’s position that OpenAI and Anthropic do not need to rely on regulation to slow the development of frontier models, while exploring the benefits, concerns, and implications for competition in AI.
Sacks believes that competition and corporate responsibility should drive development, rather than using regulation to slow progress. The benefit is that AI could advance faster and create opportunities for new innovations.
The risk, however, is that companies may prioritize winning the market over safety. Without a clear regulatory framework, users and society may bear the consequences of models that remain difficult to control. The key issue is therefore not “stop or release,” but finding a balance between competition, safety, and accountability l
This article analyzes David Sacks’s position that OpenAI and Anthropic do not need to rely on regulation to slow the development of frontier models, while exploring the benefits, concerns, and implications for competition in AI.
Sacks believes that competition and corporate responsibility should drive development, rather than using regulation to slow progress. The benefit is that AI could advance faster and create opportunities for new innovations.
The risk, however, is that companies may prioritize winning the market over safety. Without a clear regulatory framework, users and society may bear the consequences of models that remain difficult to control. The key issue is therefore not “stop or release,” but finding a balance between competition, safety, and accountability l
The Key Point Sacks Is Arguing
Sacks is arguing against using regulation as a tool to “slow down” the development of frontier models by OpenAI and Anthropic. He believes that the pace of development should not be limited by rules designed to suppress competition or innovation.
However, this does not mean that he opposes all oversight. Rules governing safety, transparency, and accountability are separate from rules whose goal is to make companies develop AI more slowly l
The Key Point Sacks Is Arguing
Sacks is arguing against using regulation as a tool to “slow down” the development of frontier models by OpenAI and Anthropic. He believes that the pace of development should not be limited by rules designed to suppress competition or innovation.
However, this does not mean that he opposes all oversight. Rules governing safety, transparency, and accountability are separate from rules whose goal is to make companies develop AI more slowly l
Who Is David Sacks, and Why Does His Opinion Carry Weight?
David Sacks is a former PayPal executive, the founder of Yammer, and a co-founder of the venture capital firm Craft Ventures. He is now known for serving as the White House’s special adviser on AI and cryptocurrency, as well as co-hosting the All-In podcast (White House)
His views therefore come not only from an investor’s perspective, but also connect Silicon Valley, capital, and government policy. When Sacks supports continued development of frontier models, his position signals to AI builders and investors that the U.S. government may prioritize competition over limiting the pace of development. At the same time, regulators must clearly weigh innovation against safety l
Who Is David Sacks, and Why Does His Opinion Carry Weight?
David Sacks is a former PayPal executive, the founder of Yammer, and a co-founder of the venture capital firm Craft Ventures. He is now known for serving as the White House’s special adviser on AI and cryptocurrency, as well as co-hosting the All-In podcast (White House)
His views therefore come not only from an investor’s perspective, but also connect Silicon Valley, capital, and government policy. When Sacks supports continued development of frontier models, his position signals to AI builders and investors that the U.S. government may prioritize competition over limiting the pace of development. At the same time, regulators must clearly weigh innovation against safety l
When AI Progress Is Seen as Too Fast
For users, new models are released so continuously that it is difficult to keep up with their capabilities, pricing, and methods of use. As AI is increasingly applied to real-world tasks, the risks of inaccurate information, privacy violations, and misuse also increase.
But if regulation becomes too strict, developers may face so much time and cost that competition slows down, especially for small teams with limited resources. Innovation could become concentrated among large companies, while users and society still need clear rules that help reduce risk l
When AI Progress Is Seen as Too Fast
For users, new models are released so continuously that it is difficult to keep up with their capabilities, pricing, and methods of use. As AI is increasingly applied to real-world tasks, the risks of inaccurate information, privacy violations, and misuse also increase.
But if regulation becomes too strict, developers may face so much time and cost that competition slows down, especially for small teams with limited resources. Innovation could become concentrated among large companies, while users and society still need clear rules that help reduce risk l
Where This Position Fits in the Broader AI Policy Landscape
Sacks’s position is on the side that does not want the government to use rules to force OpenAI and Anthropic to slow the development of frontier models, because he believes competition and innovation should continue moving forward.
This approach differs from those who support strict controls or propose pausing development to carefully assess risks first. That camp places greater importance on preventing harm than on the speed of technological progress.
Another option is to establish safety standards, such as testing, risk disclosure, and accountability to users, without imposing a speed limit. This approach seeks to maintain balance: companies can continue developing, but must demonstrate that the technology does not create unnecessary additional risks.
Where This Position Fits in the Broader AI Policy Landscape
Sacks’s position is on the side that does not want the government to use rules to force OpenAI and Anthropic to slow the development of frontier models, because he believes competition and innovation should continue moving forward.
This approach differs from those who support strict controls or propose pausing development to carefully assess risks first. That camp places greater importance on preventing harm than on the speed of technological progress.
Another option is to establish safety standards, such as testing, risk disclosure, and accountability to users, without imposing a speed limit. This approach seeks to maintain balance: companies can continue developing, but must demonstrate that the technology does not create unnecessary additional risks.
From Stopping Models to Controlling How They Are Used
From Sacks’s perspective, the key issue is not setting a speed limit for models, but regulating how companies use the technology, from launch through accountability for its impacts.
| Factor | Rules to slow development | Rules to control usage |
|---|---|---|
| Development | Limit the speed and size of models | Allow continued development under conditions |
| Launch | Delay or postpone launches | Set criteria before making models available |
| Safety testing | May not be the primary focus | Require testing and risk disclosure |
| Accountability | Focus on controlling developers | Companies are accountable for usage |
The latter approach is like placing a guardrail around real-world use. It can control risks without closing off the path to developing new models.
From Stopping Models to Controlling How They Are Used
From Sacks’s perspective, the key issue is not setting a speed limit for models, but regulating how companies use the technology, from launch through accountability for its impacts.
| Factor | Rules to slow development | Rules to control usage |
|---|---|---|
| Development | Limit the speed and size of models | Allow continued development under conditions |
| Launch | Delay or postpone launches | Set criteria before making models available |
| Safety testing | May not be the primary focus | Require testing and risk disclosure |
| Accountability | Focus on controlling developers | Companies are accountable for usage |
The latter approach is like placing a guardrail around real-world use. It can control risks without closing off the path to developing new models.
What Does “No Need to Slow Down” Mean in Practice?
This position may mean reducing pre-launch restrictions so that OpenAI and Anthropic can release models more quickly, while still holding companies accountable for risks arising from real-world use.
It may also mean accelerating competition and preventing unclear rules from becoming a shared barrier. Safety standards could evolve alongside the technology through testing, risk disclosure, and adjustments based on new information.
The key point is that “no need to slow down” does not mean that safety is unnecessary. It means companies should not have to wait for perfect rules before continuing development.
What Does “No Need to Slow Down” Mean in Practice?
This position may mean reducing pre-launch restrictions so that OpenAI and Anthropic can release models more quickly, while still holding companies accountable for risks arising from real-world use.
It may also mean accelerating competition and preventing unclear rules from becoming a shared barrier. Safety standards could evolve alongside the technology through testing, risk disclosure, and adjustments based on new information.
The key point is that “no need to slow down” does not mean that safety is unnecessary. It means companies should not have to wait for perfect rules before continuing development.
What Would Happen If OpenAI and Anthropic Were Allowed to Move Forward at Full Speed?
Automation would take on more repetitive work, from summarizing documents and analyzing data to helping write code. Researchers could use models to find information and develop hypotheses more quickly, although humans would still need to review the results.
In the military and cybersecurity fields, models could help identify vulnerabilities or respond to attacks more quickly. At the same time, malicious actors could use them to create more sophisticated attacks.
The labor market would shift from the complete replacement of jobs toward reducing the number of people needed in certain stages and increasing the importance of skills used to review, make decisions, and take responsibility for outcomes. Companies that fail to adapt could quickly lose their competitive advantage.
What Would Happen If OpenAI and Anthropic Were Allowed to Move Forward at Full Speed?
Automation would take on more repetitive work, from summarizing documents and analyzing data to helping write code. Researchers could use models to find information and develop hypotheses more quickly, although humans would still need to review the results.
In the military and cybersecurity fields, models could help identify vulnerabilities or respond to attacks more quickly. At the same time, malicious actors could use them to create more sophisticated attacks.
The labor market would shift from the complete replacement of jobs toward reducing the number of people needed in certain stages and increasing the importance of skills used to review, make decisions, and take responsibility for outcomes. Companies that fail to adapt could quickly lose their competitive advantage.
Who Benefits, and Who Bears the Risks?
Requirements can reduce the risks of models being released faster than they can be properly reviewed, but they may also make it harder for smaller players to keep up with the market. Without requirements, innovation moves faster, but more of the risk burden falls on users and society.
| Factor | With requirements | Without requirements |
|---|---|---|
| Large AI companies | Clearer review, but higher costs | Faster development, but greater risk of harm |
| Startups | Harder to enter the market | More able to compete through speed |
| Developers | Safer usage guidelines | Greater freedom to experiment |
| Consumers | Lower risk, but features may arrive later | Access to new products sooner, but must bear the risks |
| Workers | More time to reskill | Job transitions may happen faster |
| Government | More systematic oversight | Must deal with harm after it occurs |
Who Benefits, and Who Bears the Risks?
Requirements can reduce the risks of models being released faster than they can be properly reviewed, but they may also make it harder for smaller players to keep up with the market. Without requirements, innovation moves faster, but more of the risk burden falls on users and society.
| Factor | With requirements | Without requirements |
|---|---|---|
| Large AI companies | Clearer review, but higher costs | Faster development, but greater risk of harm |
| Startups | Harder to enter the market | More able to compete through speed |
| Developers | Safer usage guidelines | Greater freedom to experiment |
| Consumers | Lower risk, but features may arrive later | Access to new products sooner, but must bear the risks |
| Workers | More time to reskill | Job transitions may happen faster |
| Government | More systematic oversight | Must deal with harm after it occurs |
Benefits of Not Using Regulation to Slow Models
Not forcing OpenAI and Anthropic to slow their models helps preserve technological leadership and allows companies to compete in developing new capabilities more quickly, much like accelerating the readiness of 3nm chips for real-world use instead of waiting through procedures that could slow the market.
Startups also avoid excessive documentation and review costs, giving them more opportunities to test new ideas and create innovations that reduce risks more quickly.
Pros
- +Preserves technological leadership
- +Helps competition and innovation move faster
Cons
- −Startups may have to bear more risk
- −Solutions may lag behind model launches
Benefits of Not Using Regulation to Slow Models
Not forcing OpenAI and Anthropic to slow their models helps preserve technological leadership and allows companies to compete in developing new capabilities more quickly, much like accelerating the readiness of 3nm chips for real-world use instead of waiting through procedures that could slow the market.
Startups also avoid excessive documentation and review costs, giving them more opportunities to test new ideas and create innovations that reduce risks more quickly.
Pros
- +Preserves technological leadership
- +Helps competition and innovation move faster
Cons
- −Startups may have to bear more risk
- −Solutions may lag behind model launches
The Weakness of Relying on Companies
Allowing companies to set their own development pace may cause business incentives to outweigh safety. Transparency is also difficult to assess because important information remains in companies’ hands.
Faster competition may encourage companies to reduce testing procedures. At the same time, power and resources remain concentrated among a few companies, forcing society to place too much trust in the same established players.
Pros
- +Models can be developed and decisions made quickly
- +Companies have substantial resources to maintain systems
Cons
- −Business incentives may conflict with safety
- −Transparency and accountability are difficult to assess
The Weakness of Relying on Companies
Allowing companies to set their own development pace may cause business incentives to outweigh safety. Transparency is also difficult to assess because important information remains in companies’ hands.
Faster competition may encourage companies to reduce testing procedures. At the same time, power and resources remain concentrated among a few companies, forcing society to place too much trust in the same established players.
Pros
- +Models can be developed and decisions made quickly
- +Companies have substantial resources to maintain systems
Cons
- −Business incentives may conflict with safety
- −Transparency and accountability are difficult to assess
Even without fines or licensing requirements, the costs do not disappear. They may instead be shifted onto displaced workers, people who receive inaccurate information, and users who encounter scams involving AI-generated content.
There are also costs related to energy, data centers, and infrastructure that society must share. When systems fail, the harm rarely falls on companies alone; it spreads to consumers, organizations, and governments.
Even without fines or licensing requirements, the costs do not disappear. They may instead be shifted onto displaced workers, people who receive inaccurate information, and users who encounter scams involving AI-generated content.
There are also costs related to energy, data centers, and infrastructure that society must share. When systems fail, the harm rarely falls on companies alone; it spreads to consumers, organizations, and governments.
The Line Between Accelerating Innovation and Releasing Risk
Rules that require companies to obtain permission every time they develop something, or that prescribe a fixed research method, may create unnecessary obstacles if they do not clearly reduce risk. The key question is whether a rule is tied to a real risk or merely adds cumbersome procedures.
However, pre-release safety testing, documenting what models can do, reporting failures, and providing channels for affected people to seek explanations or remedies are all necessary. Rules of this kind do not obstruct innovation; they help companies take responsibility for real-world impacts.
The Line Between Accelerating Innovation and Releasing Risk
Rules that require companies to obtain permission every time they develop something, or that prescribe a fixed research method, may create unnecessary obstacles if they do not clearly reduce risk. The key question is whether a rule is tied to a real risk or merely adds cumbersome procedures.
However, pre-release safety testing, documenting what models can do, reporting failures, and providing channels for affected people to seek explanations or remedies are all necessary. Rules of this kind do not obstruct innovation; they help companies take responsibility for real-world impacts.
Questions to Ask About Sacks’s Proposal
Who decides whether a model is “safe enough”—the developing company, a government agency, or an independent assessor? If a model causes harm, who is responsible, and how can affected people request an explanation or remedy?
Another question is how much the public should be able to scrutinize AI companies without requiring them to disclose all their confidential business information. At minimum, testing evidence, incident reports, and the reasons behind a model’s release decision should be available for review to some extent.
Questions to Ask About Sacks’s Proposal
Who decides whether a model is “safe enough”—the developing company, a government agency, or an independent assessor? If a model causes harm, who is responsible, and how can affected people request an explanation or remedy?
Another question is how much the public should be able to scrutinize AI companies without requiring them to disclose all their confidential business information. At minimum, testing evidence, incident reports, and the reasons behind a model’s release decision should be available for review to some extent.
Conclusion: Not Slowing Down Should Not Mean Avoiding Responsibility
David Sacks’s position suggests that OpenAI and Anthropic should be able to continue developing frontier models quickly, but speed should not become an excuse to ignore social responsibility.
The central challenge may not be choosing between “accelerate” and “stop,” but designing a system that allows companies to move quickly while maintaining testing evidence, transparency, and clearly assigned responsibility. If problems occur, they must be traceable and genuinely remedied. Only then can competition in AI continue without leaving trust behind.
Conclusion: Not Slowing Down Should Not Mean Avoiding Responsibility
David Sacks’s position suggests that OpenAI and Anthropic should be able to continue developing frontier models quickly, but speed should not become an excuse to ignore social responsibility.
The central challenge may not be choosing between “accelerate” and “stop,” but designing a system that allows companies to move quickly while maintaining testing evidence, transparency, and clearly assigned responsibility. If problems occur, they must be traceable and genuinely remedied. Only then can competition in AI continue without leaving trust behind.
The Key Point Sacks Is Arguing
Sacks is not opposed to all AI regulation. He is criticizing rules specifically designed to slow the development of frontier models. He believes that using regulation as a brake could cause domestic companies to lose their competitive speed and reduce incentives to develop new technology.
This position does not mean companies should be allowed to do anything they want. Safety, transparency, and accountability rules remain necessary; they simply should not be designed to automatically become tools for stopping progress.
The Key Point Sacks Is Arguing
Sacks is not opposed to all AI regulation. He is criticizing rules specifically designed to slow the development of frontier models. He believes that using regulation as a brake could cause domestic companies to lose their competitive speed and reduce incentives to develop new technology.
This position does not mean companies should be allowed to do anything they want. Safety, transparency, and accountability rules remain necessary; they simply should not be designed to automatically become tools for stopping progress.
David Sacks is a technology entrepreneur and investor. He previously served as an executive at PayPal and co-founded Craft Ventures. He currently serves as the AI and Crypto Czar in the U.S. government, placing him at the intersection of Silicon Valley, capital, and government policy.
His opinion carries weight because he does not view AI solely from the corporate side; he also understands business competition and regulatory mechanisms. Sacks’s statements are therefore closely watched by AI developers, investors assessing risk, and government agencies designing rules for frontier models.
David Sacks is a technology entrepreneur and investor. He previously served as an executive at PayPal and co-founded Craft Ventures. He currently serves as the AI and Crypto Czar in the U.S. government, placing him at the intersection of Silicon Valley, capital, and government policy.
His opinion carries weight because he does not view AI solely from the corporate side; he also understands business competition and regulatory mechanisms. Sacks’s statements are therefore closely watched by AI developers, investors assessing risk, and government agencies designing rules for frontier models.
When AI Progress Is Seen as Too Fast
For users, new models are released so frequently that it is difficult to keep up with their capabilities, methods of use, and constantly changing risks. Society must therefore deal with impacts that have not yet been fully assessed, ranging from inaccurate information to the misuse of AI.
But if regulation becomes too strict, companies may have to slow development, allowing competitors operating under different rules to move ahead instead. The challenge is not simply choosing whether to control AI or leave it unrestricted, but designing rules that reduce risks without blocking competition and innovation.
When AI Progress Is Seen as Too Fast
For users, new models are released so frequently that it is difficult to keep up with their capabilities, methods of use, and constantly changing risks. Society must therefore deal with impacts that have not yet been fully assessed, ranging from inaccurate information to the misuse of AI.
But if regulation becomes too strict, companies may have to slow development, allowing competitors operating under different rules to move ahead instead. The challenge is not simply choosing whether to control AI or leave it unrestricted, but designing rules that reduce risks without blocking competition and innovation.
Where This Position Fits in the Broader AI Policy Landscape
Sacks’s view is opposed to rules that would force OpenAI and Anthropic to slow the development of frontier models. He believes competition should continue, and that the government should not directly impose a speed limit on companies.
This position differs from strict-control approaches or proposals to pause development, which prioritize reducing risks first. But it does not mean releasing AI without any rules, since safety standards, testing, and accountability can still be used as shared conditions.
Put plainly, this approach attempts to separate “safety” from “slowing competition.” Companies should be able to develop quickly, but they must demonstrate that their models are safe enough for real-world use.
Where This Position Fits in the Broader AI Policy Landscape
Sacks’s view is opposed to rules that would force OpenAI and Anthropic to slow the development of frontier models. He believes competition should continue, and that the government should not directly impose a speed limit on companies.
This position differs from strict-control approaches or proposals to pause development, which prioritize reducing risks first. But it does not mean releasing AI without any rules, since safety standards, testing, and accountability can still be used as shared conditions.
Put plainly, this approach attempts to separate “safety” from “slowing competition.” Companies should be able to develop quickly, but they must demonstrate that their models are safe enough for real-world use.
From Stopping Models to Controlling How They Are Used
David Sacks’s view is that rules should not be used to stop the development of frontier models. Instead, they should focus on where risks actually arise: launch, safety testing, and accountability when companies put models into use.
This approach creates room for competition while requiring companies to establish clear measures to address potential impacts.
| Factor | Slowing model development | Controlling model use |
|---|---|---|
| Primary goal | Limit the speed of model creation | Reduce risks from real-world use |
| Where rules apply | Research and development processes | Launch, testing, and usage |
| Impact on competition | May slow competition | Still allows development under conditions |
| Corporate accountability | Focus on stopping or limiting models | Require accountability for impacts |
From Stopping Models to Controlling How They Are Used
David Sacks’s view is that rules should not be used to stop the development of frontier models. Instead, they should focus on where risks actually arise: launch, safety testing, and accountability when companies put models into use.
This approach creates room for competition while requiring companies to establish clear measures to address potential impacts.
| Factor | Slowing model development | Controlling model use |
|---|---|---|
| Primary goal | Limit the speed of model creation | Reduce risks from real-world use |
| Where rules apply | Research and development processes | Launch, testing, and usage |
| Impact on competition | May slow competition | Still allows development under conditions |
| Corporate accountability | Focus on stopping or limiting models | Require accountability for impacts |
What Does “No Need to Slow Down” Mean in Practice?
This position may mean reducing pre-launch restrictions so that companies can test and release models more quickly, while using risk assessments and post-launch accountability as controls instead.
It may also mean accelerating competition between OpenAI and Anthropic, since neither company would have to wait for an overly strict regulatory framework. Innovation could therefore continue, but companies would face greater pressure to prove safety themselves.
Allowing safety standards to evolve alongside the technology means that rules should not lock development methods in advance. However, testing criteria, risk disclosure, and accountability mechanisms must also adapt to the capabilities of the models.
What Does “No Need to Slow Down” Mean in Practice?
This position may mean reducing pre-launch restrictions so that companies can test and release models more quickly, while using risk assessments and post-launch accountability as controls instead.
It may also mean accelerating competition between OpenAI and Anthropic, since neither company would have to wait for an overly strict regulatory framework. Innovation could therefore continue, but companies would face greater pressure to prove safety themselves.
Allowing safety standards to evolve alongside the technology means that rules should not lock development methods in advance. However, testing criteria, risk disclosure, and accountability mechanisms must also adapt to the capabilities of the models.
What Would Happen If OpenAI and Anthropic Were Allowed to Move Forward at Full Speed?
If OpenAI and Anthropic move forward without waiting for rules to control their pace, models will take on deeper roles in automation, research, and programming. Companies may create tools that analyze data or solve complex problems faster than before.
In national security, the same capabilities could be used both to defend and attack cyber systems, as well as to support military operations. This is therefore not merely a matter of convenience for ordinary users; it also involves risks at the national level.
The labor market will change accordingly. Some types of work may decline, while people who know how to use models will gain an advantage. The answer is not to halt development, but to provide testing, risk disclosure, and preparation to help people adapt in time.
What Would Happen If OpenAI and Anthropic Were Allowed to Move Forward at Full Speed?
If OpenAI and Anthropic move forward without waiting for rules to control their pace, models will take on deeper roles in automation, research, and programming. Companies may create tools that analyze data or solve complex problems faster than before.
In national security, the same capabilities could be used both to defend and attack cyber systems, as well as to support military operations. This is therefore not merely a matter of convenience for ordinary users; it also involves risks at the national level.
The labor market will change accordingly. Some types of work may decline, while people who know how to use models will gain an advantage. The answer is not to halt development, but to provide testing, risk disclosure, and preparation to help people adapt in time.
Who Benefits, and Who Bears the Risks?
| Factor | With speed-control requirements | Without speed-control requirements |
|---|---|---|
| Large AI companies | More time to test and reduce risks | Can release models quickly, but risks accumulate |
| Startups | Higher compliance costs | Greater flexibility to compete |
| Developers | Clearer usage guidelines | Faster access to new technology |
| Consumers | More thoroughly reviewed services | New features arrive sooner, but users bear the risks themselves |
| Workers | More time to reskill | Job transitions may happen faster |
| Government | Easier oversight | Must deal with subsequent impacts |
The key point is that requirements do not help only the government; they also reduce the shock to users and workers. At the same time, if rules are too detailed or slow, startups may lose opportunities to compete and innovation may become concentrated among large companies instead.
Who Benefits, and Who Bears the Risks?
| Factor | With speed-control requirements | Without speed-control requirements |
|---|---|---|
| Large AI companies | More time to test and reduce risks | Can release models quickly, but risks accumulate |
| Startups | Higher compliance costs | Greater flexibility to compete |
| Developers | Clearer usage guidelines | Faster access to new technology |
| Consumers | More thoroughly reviewed services | New features arrive sooner, but users bear the risks themselves |
| Workers | More time to reskill | Job transitions may happen faster |
| Government | Easier oversight | Must deal with subsequent impacts |
The key point is that requirements do not help only the government; they also reduce the shock to users and workers. At the same time, if rules are too detailed or slow, startups may lose opportunities to compete and innovation may become concentrated among large companies instead.
Benefits of Not Using Regulation to Slow Models
Allowing companies to develop quickly helps preserve leadership and keeps competition moving forward, much like the Apple A19 Pro chip (3 nm) and 12GB of RAM make it easier for phones to handle demanding tasks. Users therefore have a chance to access new technology sooner.
For startups, avoiding complex rules from the outset reduces the cost and time required to enter the market. Innovations from many teams have a greater chance of being tested in practice instead of being limited to a few large companies.
Faster development may also lead to tools that reduce risks themselves, such as more accurate monitoring systems or safer models. The benefit lies in allowing technology to solve problems as it grows.
Pros
- +Preserves technological leadership
- +Creates opportunities for startups to compete and innovate
Cons
- −Risks must be monitored throughout development
- −Impacts may occur before new rules are established
Benefits of Not Using Regulation to Slow Models
Allowing companies to develop quickly helps preserve leadership and keeps competition moving forward, much like the Apple A19 Pro chip (3 nm) and 12GB of RAM make it easier for phones to handle demanding tasks. Users therefore have a chance to access new technology sooner.
For startups, avoiding complex rules from the outset reduces the cost and time required to enter the market. Innovations from many teams have a greater chance of being tested in practice instead of being limited to a few large companies.
Faster development may also lead to tools that reduce risks themselves, such as more accurate monitoring systems or safer models. The benefit lies in allowing technology to solve problems as it grows.
Pros
- +Preserves technological leadership
- +Creates opportunities for startups to compete and innovate
Cons
- −Risks must be monitored throughout development
- −Impacts may occur before new rules are established
The Weakness of Relying on Companies
Companies have incentives to accelerate the release of frontier models to preserve their business advantages, while the details of safety testing are often difficult for outsiders to verify. Competition may therefore push some procedures down the list of priorities.
When power and information are concentrated among a few companies, society has an even harder time fully assessing the risks. Relying on self-regulation therefore comes at the cost of transparency and oversight mechanisms that remain unclear.
Pros
- +Reduces regulatory burdens that could slow innovation
- +Gives companies room to experiment with new safety approaches
Cons
- −Business incentives may conflict with safety
- −Power and information are concentrated among a few companies
The Weakness of Relying on Companies
Companies have incentives to accelerate the release of frontier models to preserve their business advantages, while the details of safety testing are often difficult for outsiders to verify. Competition may therefore push some procedures down the list of priorities.
When power and information are concentrated among a few companies, society has an even harder time fully assessing the risks. Relying on self-regulation therefore comes at the cost of transparency and oversight mechanisms that remain unclear.
Pros
- +Reduces regulatory burdens that could slow innovation
- +Gives companies room to experiment with new safety approaches
Cons
- −Business incentives may conflict with safety
- −Power and information are concentrated among a few companies
What Society Pays Even Without Fines or Licensing
The costs do not end with companies’ research budgets. Some workers may be displaced, while inaccurate information generated by AI creates correction costs, harm, and distrust that others must deal with.
Scams may also become easier to carry out, while electricity and infrastructure must expand to support rising usage. These burdens may fall on consumers, organizations, and governments, even if companies do not pay fines or apply for licenses.
Therefore, allowing companies to regulate themselves does not mean there are no costs. It simply moves those costs from corporate balance sheets onto society.
What Society Pays Even Without Fines or Licensing
The costs do not end with companies’ research budgets. Some workers may be displaced, while inaccurate information generated by AI creates correction costs, harm, and distrust that others must deal with.
Scams may also become easier to carry out, while electricity and infrastructure must expand to support rising usage. These burdens may fall on consumers, organizations, and governments, even if companies do not pay fines or apply for licenses.
Therefore, allowing companies to regulate themselves does not mean there are no costs. It simply moves those costs from corporate balance sheets onto society.
The Line Between Accelerating Innovation and Releasing Risk
Rules that require companies to obtain permission every time they develop a model may slow experimentation without genuinely improving safety. Such requirements should be reconsidered if they are not tied to verifiable risks.
However, rules concerning pre-release testing, maintaining records of failures, reporting impacts, and providing avenues for redress should be mandatory. They ensure that companies take responsibility for what they release. If a model causes harm, affected people should have the right to request explanations, correct information, and seek compensation.
The dividing line is therefore not “rules versus no rules,” but whether a rule genuinely reduces risk or merely adds procedures that slow development.
The Line Between Accelerating Innovation and Releasing Risk
Rules that require companies to obtain permission every time they develop a model may slow experimentation without genuinely improving safety. Such requirements should be reconsidered if they are not tied to verifiable risks.
However, rules concerning pre-release testing, maintaining records of failures, reporting impacts, and providing avenues for redress should be mandatory. They ensure that companies take responsibility for what they release. If a model causes harm, affected people should have the right to request explanations, correct information, and seek compensation.
The dividing line is therefore not “rules versus no rules,” but whether a rule genuinely reduces risk or merely adds procedures that slow development.
Questions to Ask About Sacks’s Proposal
Who decides whether a model is “safe enough,” and what criteria are used? If companies assess themselves, how will overlooked risks be detected?
When AI causes harm, who is responsible—the company, the developer, or the user? How much should the public be entitled to inspect testing results, risk information, and the reasons a company chose to release a model? The speed of development should not be considered more important than transparency and the rights of affected people.
Questions to Ask About Sacks’s Proposal
Who decides whether a model is “safe enough,” and what criteria are used? If companies assess themselves, how will overlooked risks be detected?
When AI causes harm, who is responsible—the company, the developer, or the user? How much should the public be entitled to inspect testing results, risk information, and the reasons a company chose to release a model? The speed of development should not be considered more important than transparency and the rights of affected people.
Conclusion: Not Slowing Down Should Not Mean Avoiding Responsibility
Accelerating the development of frontier models does not necessarily mean allowing companies to make decisions alone. It should go hand in hand with verifiable evidence, risk disclosure, and clearly identified responsibility.
The challenge may therefore not be choosing between “accelerate” and “stop,” but designing a governance system that allows companies to move quickly while requiring them to answer for impacts when they occur. Speed should go together with transparency, not replace accountability.
Conclusion: Not Slowing Down Should Not Mean Avoiding Responsibility
Accelerating the development of frontier models does not necessarily mean allowing companies to make decisions alone. It should go hand in hand with verifiable evidence, risk disclosure, and clearly identified responsibility.
The challenge may therefore not be choosing between “accelerate” and “stop,” but designing a governance system that allows companies to move quickly while requiring them to answer for impacts when they occur. Speed should go together with transparency, not replace accountability.
This article analyzes David Sacks’s position that OpenAI and Anthropic do not need to rely on regulation to slow the development of frontier models, while exploring the benefits, concerns, and implications for competition in AI.
Sacks believes that competition and corporate responsibility should drive development, rather than using regulation to slow progress. The benefit is that AI could advance faster and create opportunities for new innovations.
The risk, however, is that companies may prioritize winning the market over safety. Without a clear regulatory framework, users and society may bear the consequences of models that remain difficult to control. The key issue is therefore not “stop or release,” but finding a balance between competition, safety, and accountability l
This article analyzes David Sacks’s position that OpenAI and Anthropic do not need to rely on regulation to slow the development of frontier models, while exploring the benefits, concerns, and implications for competition in AI.
Sacks believes that competition and corporate responsibility should drive development, rather than using regulation to slow progress. The benefit is that AI could advance faster and create opportunities for new innovations.
The risk, however, is that companies may prioritize winning the market over safety. Without a clear regulatory framework, users and society may bear the consequences of models that remain difficult to control. The key issue is therefore not “stop or release,” but finding a balance between competition, safety, and accountability l
The Key Point Sacks Is Arguing
Sacks is arguing against using regulation as a tool to “slow down” the development of frontier models by OpenAI and Anthropic. He believes that the pace of development should not be limited by rules designed to suppress competition or innovation.
However, this does not mean that he opposes all oversight. Rules governing safety, transparency, and accountability are separate from rules whose goal is to make companies develop AI more slowly l
The Key Point Sacks Is Arguing
Sacks is arguing against using regulation as a tool to “slow down” the development of frontier models by OpenAI and Anthropic. He believes that the pace of development should not be limited by rules designed to suppress competition or innovation.
However, this does not mean that he opposes all oversight. Rules governing safety, transparency, and accountability are separate from rules whose goal is to make companies develop AI more slowly l
Who Is David Sacks, and Why Does His Opinion Carry Weight?
David Sacks is a former PayPal executive, the founder of Yammer, and a co-founder of the venture capital firm Craft Ventures. He is now known for serving as the White House’s special adviser on AI and cryptocurrency, as well as co-hosting the All-In podcast (White House)
His views therefore come not only from an investor’s perspective, but also connect Silicon Valley, capital, and government policy. When Sacks supports continued development of frontier models, his position signals to AI builders and investors that the U.S. government may prioritize competition over limiting the pace of development. At the same time, regulators must clearly weigh innovation against safety l
Who Is David Sacks, and Why Does His Opinion Carry Weight?
David Sacks is a former PayPal executive, the founder of Yammer, and a co-founder of the venture capital firm Craft Ventures. He is now known for serving as the White House’s special adviser on AI and cryptocurrency, as well as co-hosting the All-In podcast (White House)
His views therefore come not only from an investor’s perspective, but also connect Silicon Valley, capital, and government policy. When Sacks supports continued development of frontier models, his position signals to AI builders and investors that the U.S. government may prioritize competition over limiting the pace of development. At the same time, regulators must clearly weigh innovation against safety l
When AI Progress Is Seen as Too Fast
For users, new models are released so continuously that it is difficult to keep up with their capabilities, pricing, and methods of use. As AI is increasingly applied to real-world tasks, the risks of inaccurate information, privacy violations, and misuse also increase.
But if regulation becomes too strict, developers may face so much time and cost that competition slows down, especially for small teams with limited resources. Innovation could become concentrated among large companies, while users and society still need clear rules that help reduce risk l
When AI Progress Is Seen as Too Fast
For users, new models are released so continuously that it is difficult to keep up with their capabilities, pricing, and methods of use. As AI is increasingly applied to real-world tasks, the risks of inaccurate information, privacy violations, and misuse also increase.
But if regulation becomes too strict, developers may face so much time and cost that competition slows down, especially for small teams with limited resources. Innovation could become concentrated among large companies, while users and society still need clear rules that help reduce risk l
Where This Position Fits in the Broader AI Policy Landscape
Sacks’s position is on the side that does not want the government to use rules to force OpenAI and Anthropic to slow the development of frontier models, because he believes competition and innovation should continue moving forward.
This approach differs from those who support strict controls or propose pausing development to carefully assess risks first. That camp places greater importance on preventing harm than on the speed of technological progress.
Another option is to establish safety standards, such as testing, risk disclosure, and accountability to users, without imposing a speed limit. This approach seeks to maintain balance: companies can continue developing, but must demonstrate that the technology does not create unnecessary additional risks.
Where This Position Fits in the Broader AI Policy Landscape
Sacks’s position is on the side that does not want the government to use rules to force OpenAI and Anthropic to slow the development of frontier models, because he believes competition and innovation should continue moving forward.
This approach differs from those who support strict controls or propose pausing development to carefully assess risks first. That camp places greater importance on preventing harm than on the speed of technological progress.
Another option is to establish safety standards, such as testing, risk disclosure, and accountability to users, without imposing a speed limit. This approach seeks to maintain balance: companies can continue developing, but must demonstrate that the technology does not create unnecessary additional risks.
From Stopping Models to Controlling How They Are Used
From Sacks’s perspective, the key issue is not setting a speed limit for models, but regulating how companies use the technology, from launch through accountability for its impacts.
| Factor | Rules to slow development | Rules to control usage |
|---|---|---|
| Development | Limit the speed and size of models | Allow continued development under conditions |
| Launch | Delay or postpone launches | Set criteria before making models available |
| Safety testing | May not be the primary focus | Require testing and risk disclosure |
| Accountability | Focus on controlling developers | Companies are accountable for usage |
The latter approach is like placing a guardrail around real-world use. It can control risks without closing off the path to developing new models.
From Stopping Models to Controlling How They Are Used
From Sacks’s perspective, the key issue is not setting a speed limit for models, but regulating how companies use the technology, from launch through accountability for its impacts.
| Factor | Rules to slow development | Rules to control usage |
|---|---|---|
| Development | Limit the speed and size of models | Allow continued development under conditions |
| Launch | Delay or postpone launches | Set criteria before making models available |
| Safety testing | May not be the primary focus | Require testing and risk disclosure |
| Accountability | Focus on controlling developers | Companies are accountable for usage |
The latter approach is like placing a guardrail around real-world use. It can control risks without closing off the path to developing new models.
What Does “No Need to Slow Down” Mean in Practice?
This position may mean reducing pre-launch restrictions so that OpenAI and Anthropic can release models more quickly, while still holding companies accountable for risks arising from real-world use.
It may also mean accelerating competition and preventing unclear rules from becoming a shared barrier. Safety standards could evolve alongside the technology through testing, risk disclosure, and adjustments based on new information.
The key point is that “no need to slow down” does not mean that safety is unnecessary. It means companies should not have to wait for perfect rules before continuing development.
What Does “No Need to Slow Down” Mean in Practice?
This position may mean reducing pre-launch restrictions so that OpenAI and Anthropic can release models more quickly, while still holding companies accountable for risks arising from real-world use.
It may also mean accelerating competition and preventing unclear rules from becoming a shared barrier. Safety standards could evolve alongside the technology through testing, risk disclosure, and adjustments based on new information.
The key point is that “no need to slow down” does not mean that safety is unnecessary. It means companies should not have to wait for perfect rules before continuing development.
What Would Happen If OpenAI and Anthropic Were Allowed to Move Forward at Full Speed?
Automation would take on more repetitive work, from summarizing documents and analyzing data to helping write code. Researchers could use models to find information and develop hypotheses more quickly, although humans would still need to review the results.
In the military and cybersecurity fields, models could help identify vulnerabilities or respond to attacks more quickly. At the same time, malicious actors could use them to create more sophisticated attacks.
The labor market would shift from the complete replacement of jobs toward reducing the number of people needed in certain stages and increasing the importance of skills used to review, make decisions, and take responsibility for outcomes. Companies that fail to adapt could quickly lose their competitive advantage.
What Would Happen If OpenAI and Anthropic Were Allowed to Move Forward at Full Speed?
Automation would take on more repetitive work, from summarizing documents and analyzing data to helping write code. Researchers could use models to find information and develop hypotheses more quickly, although humans would still need to review the results.
In the military and cybersecurity fields, models could help identify vulnerabilities or respond to attacks more quickly. At the same time, malicious actors could use them to create more sophisticated attacks.
The labor market would shift from the complete replacement of jobs toward reducing the number of people needed in certain stages and increasing the importance of skills used to review, make decisions, and take responsibility for outcomes. Companies that fail to adapt could quickly lose their competitive advantage.
Who Benefits, and Who Bears the Risks?
Requirements can reduce the risks of models being released faster than they can be properly reviewed, but they may also make it harder for smaller players to keep up with the market. Without requirements, innovation moves faster, but more of the risk burden falls on users and society.
| Factor | With requirements | Without requirements |
|---|---|---|
| Large AI companies | Clearer review, but higher costs | Faster development, but greater risk of harm |
| Startups | Harder to enter the market | More able to compete through speed |
| Developers | Safer usage guidelines | Greater freedom to experiment |
| Consumers | Lower risk, but features may arrive later | Access to new products sooner, but must bear the risks |
| Workers | More time to reskill | Job transitions may happen faster |
| Government | More systematic oversight | Must deal with harm after it occurs |
Who Benefits, and Who Bears the Risks?
Requirements can reduce the risks of models being released faster than they can be properly reviewed, but they may also make it harder for smaller players to keep up with the market. Without requirements, innovation moves faster, but more of the risk burden falls on users and society.
| Factor | With requirements | Without requirements |
|---|---|---|
| Large AI companies | Clearer review, but higher costs | Faster development, but greater risk of harm |
| Startups | Harder to enter the market | More able to compete through speed |
| Developers | Safer usage guidelines | Greater freedom to experiment |
| Consumers | Lower risk, but features may arrive later | Access to new products sooner, but must bear the risks |
| Workers | More time to reskill | Job transitions may happen faster |
| Government | More systematic oversight | Must deal with harm after it occurs |
Benefits of Not Using Regulation to Slow Models
Not forcing OpenAI and Anthropic to slow their models helps preserve technological leadership and allows companies to compete in developing new capabilities more quickly, much like accelerating the readiness of 3nm chips for real-world use instead of waiting through procedures that could slow the market.
Startups also avoid excessive documentation and review costs, giving them more opportunities to test new ideas and create innovations that reduce risks more quickly.
Pros
- +Preserves technological leadership
- +Helps competition and innovation move faster
Cons
- −Startups may have to bear more risk
- −Solutions may lag behind model launches
Benefits of Not Using Regulation to Slow Models
Not forcing OpenAI and Anthropic to slow their models helps preserve technological leadership and allows companies to compete in developing new capabilities more quickly, much like accelerating the readiness of 3nm chips for real-world use instead of waiting through procedures that could slow the market.
Startups also avoid excessive documentation and review costs, giving them more opportunities to test new ideas and create innovations that reduce risks more quickly.
Pros
- +Preserves technological leadership
- +Helps competition and innovation move faster
Cons
- −Startups may have to bear more risk
- −Solutions may lag behind model launches
The Weakness of Relying on Companies
Allowing companies to set their own development pace may cause business incentives to outweigh safety. Transparency is also difficult to assess because important information remains in companies’ hands.
Faster competition may encourage companies to reduce testing procedures. At the same time, power and resources remain concentrated among a few companies, forcing society to place too much trust in the same established players.
Pros
- +Models can be developed and decisions made quickly
- +Companies have substantial resources to maintain systems
Cons
- −Business incentives may conflict with safety
- −Transparency and accountability are difficult to assess
The Weakness of Relying on Companies
Allowing companies to set their own development pace may cause business incentives to outweigh safety. Transparency is also difficult to assess because important information remains in companies’ hands.
Faster competition may encourage companies to reduce testing procedures. At the same time, power and resources remain concentrated among a few companies, forcing society to place too much trust in the same established players.
Pros
- +Models can be developed and decisions made quickly
- +Companies have substantial resources to maintain systems
Cons
- −Business incentives may conflict with safety
- −Transparency and accountability are difficult to assess
Even without fines or licensing requirements, the costs do not disappear. They may instead be shifted onto displaced workers, people who receive inaccurate information, and users who encounter scams involving AI-generated content.
There are also costs related to energy, data centers, and infrastructure that society must share. When systems fail, the harm rarely falls on companies alone; it spreads to consumers, organizations, and governments.
Even without fines or licensing requirements, the costs do not disappear. They may instead be shifted onto displaced workers, people who receive inaccurate information, and users who encounter scams involving AI-generated content.
There are also costs related to energy, data centers, and infrastructure that society must share. When systems fail, the harm rarely falls on companies alone; it spreads to consumers, organizations, and governments.
The Line Between Accelerating Innovation and Releasing Risk
Rules that require companies to obtain permission every time they develop something, or that prescribe a fixed research method, may create unnecessary obstacles if they do not clearly reduce risk. The key question is whether a rule is tied to a real risk or merely adds cumbersome procedures.
However, pre-release safety testing, documenting what models can do, reporting failures, and providing channels for affected people to seek explanations or remedies are all necessary. Rules of this kind do not obstruct innovation; they help companies take responsibility for real-world impacts.
The Line Between Accelerating Innovation and Releasing Risk
Rules that require companies to obtain permission every time they develop something, or that prescribe a fixed research method, may create unnecessary obstacles if they do not clearly reduce risk. The key question is whether a rule is tied to a real risk or merely adds cumbersome procedures.
However, pre-release safety testing, documenting what models can do, reporting failures, and providing channels for affected people to seek explanations or remedies are all necessary. Rules of this kind do not obstruct innovation; they help companies take responsibility for real-world impacts.
Questions to Ask About Sacks’s Proposal
Who decides whether a model is “safe enough”—the developing company, a government agency, or an independent assessor? If a model causes harm, who is responsible, and how can affected people request an explanation or remedy?
Another question is how much the public should be able to scrutinize AI companies without requiring them to disclose all their confidential business information. At minimum, testing evidence, incident reports, and the reasons behind a model’s release decision should be available for review to some extent.
Questions to Ask About Sacks’s Proposal
Who decides whether a model is “safe enough”—the developing company, a government agency, or an independent assessor? If a model causes harm, who is responsible, and how can affected people request an explanation or remedy?
Another question is how much the public should be able to scrutinize AI companies without requiring them to disclose all their confidential business information. At minimum, testing evidence, incident reports, and the reasons behind a model’s release decision should be available for review to some extent.
Conclusion: Not Slowing Down Should Not Mean Avoiding Responsibility
David Sacks’s position suggests that OpenAI and Anthropic should be able to continue developing frontier models quickly, but speed should not become an excuse to ignore social responsibility.
The central challenge may not be choosing between “accelerate” and “stop,” but designing a system that allows companies to move quickly while maintaining testing evidence, transparency, and clearly assigned responsibility. If problems occur, they must be traceable and genuinely remedied. Only then can competition in AI continue without leaving trust behind.
Conclusion: Not Slowing Down Should Not Mean Avoiding Responsibility
David Sacks’s position suggests that OpenAI and Anthropic should be able to continue developing frontier models quickly, but speed should not become an excuse to ignore social responsibility.
The central challenge may not be choosing between “accelerate” and “stop,” but designing a system that allows companies to move quickly while maintaining testing evidence, transparency, and clearly assigned responsibility. If problems occur, they must be traceable and genuinely remedied. Only then can competition in AI continue without leaving trust behind.
The Key Point Sacks Is Arguing
Sacks is not opposed to all AI regulation. He is criticizing rules specifically designed to slow the development of frontier models. He believes that using regulation as a brake could cause domestic companies to lose their competitive speed and reduce incentives to develop new technology.
This position does not mean companies should be allowed to do anything they want. Safety, transparency, and accountability rules remain necessary; they simply should not be designed to automatically become tools for stopping progress.
The Key Point Sacks Is Arguing
Sacks is not opposed to all AI regulation. He is criticizing rules specifically designed to slow the development of frontier models. He believes that using regulation as a brake could cause domestic companies to lose their competitive speed and reduce incentives to develop new technology.
This position does not mean companies should be allowed to do anything they want. Safety, transparency, and accountability rules remain necessary; they simply should not be designed to automatically become tools for stopping progress.
David Sacks is a technology entrepreneur and investor. He previously served as an executive at PayPal and co-founded Craft Ventures. He currently serves as the AI and Crypto Czar in the U.S. government, placing him at the intersection of Silicon Valley, capital, and government policy.
His opinion carries weight because he does not view AI solely from the corporate side; he also understands business competition and regulatory mechanisms. Sacks’s statements are therefore closely watched by AI developers, investors assessing risk, and government agencies designing rules for frontier models.
David Sacks is a technology entrepreneur and investor. He previously served as an executive at PayPal and co-founded Craft Ventures. He currently serves as the AI and Crypto Czar in the U.S. government, placing him at the intersection of Silicon Valley, capital, and government policy.
His opinion carries weight because he does not view AI solely from the corporate side; he also understands business competition and regulatory mechanisms. Sacks’s statements are therefore closely watched by AI developers, investors assessing risk, and government agencies designing rules for frontier models.
When AI Progress Is Seen as Too Fast
For users, new models are released so frequently that it is difficult to keep up with their capabilities, methods of use, and constantly changing risks. Society must therefore deal with impacts that have not yet been fully assessed, ranging from inaccurate information to the misuse of AI.
But if regulation becomes too strict, companies may have to slow development, allowing competitors operating under different rules to move ahead instead. The challenge is not simply choosing whether to control AI or leave it unrestricted, but designing rules that reduce risks without blocking competition and innovation.
When AI Progress Is Seen as Too Fast
For users, new models are released so frequently that it is difficult to keep up with their capabilities, methods of use, and constantly changing risks. Society must therefore deal with impacts that have not yet been fully assessed, ranging from inaccurate information to the misuse of AI.
But if regulation becomes too strict, companies may have to slow development, allowing competitors operating under different rules to move ahead instead. The challenge is not simply choosing whether to control AI or leave it unrestricted, but designing rules that reduce risks without blocking competition and innovation.
Where This Position Fits in the Broader AI Policy Landscape
Sacks’s view is opposed to rules that would force OpenAI and Anthropic to slow the development of frontier models. He believes competition should continue, and that the government should not directly impose a speed limit on companies.
This position differs from strict-control approaches or proposals to pause development, which prioritize reducing risks first. But it does not mean releasing AI without any rules, since safety standards, testing, and accountability can still be used as shared conditions.
Put plainly, this approach attempts to separate “safety” from “slowing competition.” Companies should be able to develop quickly, but they must demonstrate that their models are safe enough for real-world use.
Where This Position Fits in the Broader AI Policy Landscape
Sacks’s view is opposed to rules that would force OpenAI and Anthropic to slow the development of frontier models. He believes competition should continue, and that the government should not directly impose a speed limit on companies.
This position differs from strict-control approaches or proposals to pause development, which prioritize reducing risks first. But it does not mean releasing AI without any rules, since safety standards, testing, and accountability can still be used as shared conditions.
Put plainly, this approach attempts to separate “safety” from “slowing competition.” Companies should be able to develop quickly, but they must demonstrate that their models are safe enough for real-world use.
From Stopping Models to Controlling How They Are Used
David Sacks’s view is that rules should not be used to stop the development of frontier models. Instead, they should focus on where risks actually arise: launch, safety testing, and accountability when companies put models into use.
This approach creates room for competition while requiring companies to establish clear measures to address potential impacts.
| Factor | Slowing model development | Controlling model use |
|---|---|---|
| Primary goal | Limit the speed of model creation | Reduce risks from real-world use |
| Where rules apply | Research and development processes | Launch, testing, and usage |
| Impact on competition | May slow competition | Still allows development under conditions |
| Corporate accountability | Focus on stopping or limiting models | Require accountability for impacts |
From Stopping Models to Controlling How They Are Used
David Sacks’s view is that rules should not be used to stop the development of frontier models. Instead, they should focus on where risks actually arise: launch, safety testing, and accountability when companies put models into use.
This approach creates room for competition while requiring companies to establish clear measures to address potential impacts.
| Factor | Slowing model development | Controlling model use |
|---|---|---|
| Primary goal | Limit the speed of model creation | Reduce risks from real-world use |
| Where rules apply | Research and development processes | Launch, testing, and usage |
| Impact on competition | May slow competition | Still allows development under conditions |
| Corporate accountability | Focus on stopping or limiting models | Require accountability for impacts |
What Does “No Need to Slow Down” Mean in Practice?
This position may mean reducing pre-launch restrictions so that companies can test and release models more quickly, while using risk assessments and post-launch accountability as controls instead.
It may also mean accelerating competition between OpenAI and Anthropic, since neither company would have to wait for an overly strict regulatory framework. Innovation could therefore continue, but companies would face greater pressure to prove safety themselves.
Allowing safety standards to evolve alongside the technology means that rules should not lock development methods in advance. However, testing criteria, risk disclosure, and accountability mechanisms must also adapt to the capabilities of the models.
What Does “No Need to Slow Down” Mean in Practice?
This position may mean reducing pre-launch restrictions so that companies can test and release models more quickly, while using risk assessments and post-launch accountability as controls instead.
It may also mean accelerating competition between OpenAI and Anthropic, since neither company would have to wait for an overly strict regulatory framework. Innovation could therefore continue, but companies would face greater pressure to prove safety themselves.
Allowing safety standards to evolve alongside the technology means that rules should not lock development methods in advance. However, testing criteria, risk disclosure, and accountability mechanisms must also adapt to the capabilities of the models.
What Would Happen If OpenAI and Anthropic Were Allowed to Move Forward at Full Speed?
If OpenAI and Anthropic move forward without waiting for rules to control their pace, models will take on deeper roles in automation, research, and programming. Companies may create tools that analyze data or solve complex problems faster than before.
In national security, the same capabilities could be used both to defend and attack cyber systems, as well as to support military operations. This is therefore not merely a matter of convenience for ordinary users; it also involves risks at the national level.
The labor market will change accordingly. Some types of work may decline, while people who know how to use models will gain an advantage. The answer is not to halt development, but to provide testing, risk disclosure, and preparation to help people adapt in time.
What Would Happen If OpenAI and Anthropic Were Allowed to Move Forward at Full Speed?
If OpenAI and Anthropic move forward without waiting for rules to control their pace, models will take on deeper roles in automation, research, and programming. Companies may create tools that analyze data or solve complex problems faster than before.
In national security, the same capabilities could be used both to defend and attack cyber systems, as well as to support military operations. This is therefore not merely a matter of convenience for ordinary users; it also involves risks at the national level.
The labor market will change accordingly. Some types of work may decline, while people who know how to use models will gain an advantage. The answer is not to halt development, but to provide testing, risk disclosure, and preparation to help people adapt in time.
Who Benefits, and Who Bears the Risks?
| Factor | With speed-control requirements | Without speed-control requirements |
|---|---|---|
| Large AI companies | More time to test and reduce risks | Can release models quickly, but risks accumulate |
| Startups | Higher compliance costs | Greater flexibility to compete |
| Developers | Clearer usage guidelines | Faster access to new technology |
| Consumers | More thoroughly reviewed services | New features arrive sooner, but users bear the risks themselves |
| Workers | More time to reskill | Job transitions may happen faster |
| Government | Easier oversight | Must deal with subsequent impacts |
The key point is that requirements do not help only the government; they also reduce the shock to users and workers. At the same time, if rules are too detailed or slow, startups may lose opportunities to compete and innovation may become concentrated among large companies instead.
Who Benefits, and Who Bears the Risks?
| Factor | With speed-control requirements | Without speed-control requirements |
|---|---|---|
| Large AI companies | More time to test and reduce risks | Can release models quickly, but risks accumulate |
| Startups | Higher compliance costs | Greater flexibility to compete |
| Developers | Clearer usage guidelines | Faster access to new technology |
| Consumers | More thoroughly reviewed services | New features arrive sooner, but users bear the risks themselves |
| Workers | More time to reskill | Job transitions may happen faster |
| Government | Easier oversight | Must deal with subsequent impacts |
The key point is that requirements do not help only the government; they also reduce the shock to users and workers. At the same time, if rules are too detailed or slow, startups may lose opportunities to compete and innovation may become concentrated among large companies instead.
Benefits of Not Using Regulation to Slow Models
Allowing companies to develop quickly helps preserve leadership and keeps competition moving forward, much like the Apple A19 Pro chip (3 nm) and 12GB of RAM make it easier for phones to handle demanding tasks. Users therefore have a chance to access new technology sooner.
For startups, avoiding complex rules from the outset reduces the cost and time required to enter the market. Innovations from many teams have a greater chance of being tested in practice instead of being limited to a few large companies.
Faster development may also lead to tools that reduce risks themselves, such as more accurate monitoring systems or safer models. The benefit lies in allowing technology to solve problems as it grows.
Pros
- +Preserves technological leadership
- +Creates opportunities for startups to compete and innovate
Cons
- −Risks must be monitored throughout development
- −Impacts may occur before new rules are established
Benefits of Not Using Regulation to Slow Models
Allowing companies to develop quickly helps preserve leadership and keeps competition moving forward, much like the Apple A19 Pro chip (3 nm) and 12GB of RAM make it easier for phones to handle demanding tasks. Users therefore have a chance to access new technology sooner.
For startups, avoiding complex rules from the outset reduces the cost and time required to enter the market. Innovations from many teams have a greater chance of being tested in practice instead of being limited to a few large companies.
Faster development may also lead to tools that reduce risks themselves, such as more accurate monitoring systems or safer models. The benefit lies in allowing technology to solve problems as it grows.
Pros
- +Preserves technological leadership
- +Creates opportunities for startups to compete and innovate
Cons
- −Risks must be monitored throughout development
- −Impacts may occur before new rules are established
The Weakness of Relying on Companies
Companies have incentives to accelerate the release of frontier models to preserve their business advantages, while the details of safety testing are often difficult for outsiders to verify. Competition may therefore push some procedures down the list of priorities.
When power and information are concentrated among a few companies, society has an even harder time fully assessing the risks. Relying on self-regulation therefore comes at the cost of transparency and oversight mechanisms that remain unclear.
Pros
- +Reduces regulatory burdens that could slow innovation
- +Gives companies room to experiment with new safety approaches
Cons
- −Business incentives may conflict with safety
- −Power and information are concentrated among a few companies
The Weakness of Relying on Companies
Companies have incentives to accelerate the release of frontier models to preserve their business advantages, while the details of safety testing are often difficult for outsiders to verify. Competition may therefore push some procedures down the list of priorities.
When power and information are concentrated among a few companies, society has an even harder time fully assessing the risks. Relying on self-regulation therefore comes at the cost of transparency and oversight mechanisms that remain unclear.
Pros
- +Reduces regulatory burdens that could slow innovation
- +Gives companies room to experiment with new safety approaches
Cons
- −Business incentives may conflict with safety
- −Power and information are concentrated among a few companies
What Society Pays Even Without Fines or Licensing
The costs do not end with companies’ research budgets. Some workers may be displaced, while inaccurate information generated by AI creates correction costs, harm, and distrust that others must deal with.
Scams may also become easier to carry out, while electricity and infrastructure must expand to support rising usage. These burdens may fall on consumers, organizations, and governments, even if companies do not pay fines or apply for licenses.
Therefore, allowing companies to regulate themselves does not mean there are no costs. It simply moves those costs from corporate balance sheets onto society.
What Society Pays Even Without Fines or Licensing
The costs do not end with companies’ research budgets. Some workers may be displaced, while inaccurate information generated by AI creates correction costs, harm, and distrust that others must deal with.
Scams may also become easier to carry out, while electricity and infrastructure must expand to support rising usage. These burdens may fall on consumers, organizations, and governments, even if companies do not pay fines or apply for licenses.
Therefore, allowing companies to regulate themselves does not mean there are no costs. It simply moves those costs from corporate balance sheets onto society.
The Line Between Accelerating Innovation and Releasing Risk
Rules that require companies to obtain permission every time they develop a model may slow experimentation without genuinely improving safety. Such requirements should be reconsidered if they are not tied to verifiable risks.
However, rules concerning pre-release testing, maintaining records of failures, reporting impacts, and providing avenues for redress should be mandatory. They ensure that companies take responsibility for what they release. If a model causes harm, affected people should have the right to request explanations, correct information, and seek compensation.
The dividing line is therefore not “rules versus no rules,” but whether a rule genuinely reduces risk or merely adds procedures that slow development.
The Line Between Accelerating Innovation and Releasing Risk
Rules that require companies to obtain permission every time they develop a model may slow experimentation without genuinely improving safety. Such requirements should be reconsidered if they are not tied to verifiable risks.
However, rules concerning pre-release testing, maintaining records of failures, reporting impacts, and providing avenues for redress should be mandatory. They ensure that companies take responsibility for what they release. If a model causes harm, affected people should have the right to request explanations, correct information, and seek compensation.
The dividing line is therefore not “rules versus no rules,” but whether a rule genuinely reduces risk or merely adds procedures that slow development.
Questions to Ask About Sacks’s Proposal
Who decides whether a model is “safe enough,” and what criteria are used? If companies assess themselves, how will overlooked risks be detected?
When AI causes harm, who is responsible—the company, the developer, or the user? How much should the public be entitled to inspect testing results, risk information, and the reasons a company chose to release a model? The speed of development should not be considered more important than transparency and the rights of affected people.
Questions to Ask About Sacks’s Proposal
Who decides whether a model is “safe enough,” and what criteria are used? If companies assess themselves, how will overlooked risks be detected?
When AI causes harm, who is responsible—the company, the developer, or the user? How much should the public be entitled to inspect testing results, risk information, and the reasons a company chose to release a model? The speed of development should not be considered more important than transparency and the rights of affected people.
Conclusion: Not Slowing Down Should Not Mean Avoiding Responsibility
Accelerating the development of frontier models does not necessarily mean allowing companies to make decisions alone. It should go hand in hand with verifiable evidence, risk disclosure, and clearly identified responsibility.
The challenge may therefore not be choosing between “accelerate” and “stop,” but designing a governance system that allows companies to move quickly while requiring them to answer for impacts when they occur. Speed should go together with transparency, not replace accountability.
Conclusion: Not Slowing Down Should Not Mean Avoiding Responsibility
Accelerating the development of frontier models does not necessarily mean allowing companies to make decisions alone. It should go hand in hand with verifiable evidence, risk disclosure, and clearly identified responsibility.
The challenge may therefore not be choosing between “accelerate” and “stop,” but designing a governance system that allows companies to move quickly while requiring them to answer for impacts when they occur. Speed should go together with transparency, not replace accountability.