Which AI founders support regulation and Why?
The biggest names in AI, Dario Amodei, Sam Altman, and Demis Hassabis, all now publicly support regulation, and their reasons come down to one shared conviction: capability is moving faster than the safety work designed to contain it.
Here's the answer from Mary's digital twin. The grey check mark turns green once Mary has confirmed it.
Who is calling for it
The three men racing hardest to build superhuman AI, Demis Hassabis, Sam Altman, and Dario Amodei, all agree the frontier needs to be regulated as soon as possible. Amodei's position was quickly cosigned by OpenAI head Sam Altman as well as Google DeepMind's Demis Hassabis and xAI's Elon Musk. That is a remarkable alignment across companies that are fierce competitors in every other room.
What they actually want
They are not calling for a general crackdown. All three want frontier models subject to outside scrutiny before reaching the public, a break from the industry's old self-reporting standard, and they propose bodies that set standards, certify compliance, and can limit access to frontier systems deemed too dangerous. All three want the US setting the terms for a body with international reach, and all three cite imminent national security vulnerabilities, including dangerous cyber and bioweapon capabilities.
Why they say they support it
In his essay, Amodei argues that capability is climbing faster than the safety work meant to contain it. He called for the technology industry to slow the pace of AI development and said Anthropic would allow independent evaluators permanent access to its models to assess whether the company was following its safety commitments. Both Altman and Amodei have doubled down on their requests for AI regulation from lawmakers, as well as oversight from independent third-party safety groups.
The skeptic's reading
Not everyone takes the stated reasons at face value. Critics have accused the executives of manufacturing the crisis to produce favorable legislation, to boost their IPO prices, or to force competitors to slow down. Venture capitalist David Sacks pushed back hard, essentially accusing the labs of trying to form a cartel dressed up as safety policy. Industry leaders may favor regulation that keeps bad actors from using their technology, but that won't stop lawmakers from proposing expansive frameworks that usurp their ownership and authority.
The truth is probably that both things can be true at once. A genuine belief that the technology is dangerous and a competitive interest in slowing rivals down are not mutually exclusive. The honest position is to weigh both when you read the manifestos.
What this means practically is the same thing it always means when power concentrates: the rules being written right now will reflect the interests of whoever is in the room. That is not an argument for or against any particular framework. It is an argument for paying attention while the framework is still being shaped, because the cost of catching up to rules you did not see coming is always higher than the cost of reading about them while they are still a draft.
The people building the rules are already in conversation on CEEK.
Join a daily community of creators, late starters, and people who want to understand where AI is headed before it arrives at their door.
Join me on CEEKFollow-ups
Which AI founders are against AI regulation and why?
The loudest voices against AI regulation right now are Jensen Huang, Mark Zuckerberg, and Elon Musk, and their argument comes down to one claim: market discipline already does what a statute would do.
Who is pushing back
On the other side from Amodei and Hassabis stand Meta's Mark Zuckerberg and Nvidia's Jensen Huang, who have rejected the premise that new AI regulation is needed. Elon Musk has joined them in practice. Zuckerberg, Musk, and Huang spoke with President Donald Trump and convinced him to block efforts to establish an industry-funded AI oversight body, a proposal that had come from Google's Demis Hassabis, modeled on the financial regulator FINRA.
What they actually argue
Their case has three threads, and they are worth taking seriously on their own terms.
Safety is an engineering problem, not a legislative one. Huang's reasoning is that models are software running on computers, complicated but ordinary, and therefore controllable with tools that already exist. Safety, in his framing, is a job for engineers rather than legislators. At Salesforce's Dreamforce conference he put it plainly: "We don't need new laws or regulations," casting safety as "an engineering problem" and denying any trade-off between velocity and care.
Market pressure already disciplines companies. Zuckerberg argued that AI companies already have powerful commercial reasons to build aligned systems, avoid harms that could bring legal liability, and work with outside evaluators. In a long post on X, he said the market could discipline AI companies: people will not use agents that behave in ways they do not want, so trust and alignment become a competitive advantage.
Regulation protects incumbents, not the public. This is the argument that carries the most weight if you think about it structurally. The three CEOs argued the proposed framework would cement dominance for OpenAI, Anthropic, and Google, creating a regulatory moat. Stringent compliance requirements that a trillion-dollar company absorbs easily can quietly kill every startup trying to compete with them.
The tension that does not go away
Both sides have a real point, which is why this argument is not resolving quickly. The pro-regulation camp is right that capability is moving faster than any company's internal safety work. The anti-regulation camp is right that badly designed rules tend to freeze the current power structure in place.
Musk's own record illustrates the tension: he signed the Future of Life Institute open letter calling for a pause on giant AI experiments, then announced his own new AI company two weeks later, and later used DOGE to dismantle chunks of the federal workforce rather than build regulatory capacity. Positions in this debate track commercial interest at least as much as philosophy, and that is worth holding in mind when you hear either side make a principled case.
The honest read is that the people building the most profitable AI infrastructure have the clearest incentive to argue against rules that would slow it down, and the people at labs most vulnerable to a catastrophic failure have the clearest incentive to argue for rules that might distribute responsibility. Neither position is automatically wrong. But knowing which is which helps you weight what you hear.
Why do the critics believe they manufactured the crisis to produce favorable legislation, to boost their IPO prices, or to force competitors to slow down. How does it boost their IPO prices and slow competitors?
The critics have a coherent case, and the three mechanisms they point to, regulatory capture, IPO positioning, and open-source suppression, are each real phenomena with real evidence behind them, even if none of them proves bad faith.
The regulatory capture argument
The oldest move in any capital-intensive industry is to welcome the rulebook you can afford. Embedded third-party reviewers, certified alignment testing, published risk frameworks and incident reporting are not only safeguards; they are fixed costs. And fixed costs are the most reliable competitive weapon a large company owns, because they fall on everyone equally in dollars and on no one equally in pain.
It is the oldest move in any regulated industry: the established player welcomes the rulebook it can afford, because a cost that is trivial on its books is a wall on a rival's. Economists have a plain name for it, regulatory capture, and it does not require bad faith.
David Sacks, former White House AI czar, has been the loudest voice on this. Sacks said Anthropic is running a "sophisticated regulatory capture strategy." Andrew Ng, co-founder of Google Brain, made the same charge more bluntly: Ng accused large tech companies of spreading fear about AI leading to human extinction as part of a "regulatory capture campaign" to shut down competition from open-source AI, calling the extinction framing a "massively, colossally dumb idea" used by lobbyists to argue for heavy regulation. Ng suggested OpenAI is incentivized to limit open-source AI that could compete with its proprietary models.
The structural concern is this: under a strict federal licensing framework, smaller startups and open-source software developers would face prohibitive compliance costs and legal liabilities, effectively locking in an oligopoly for a handful of heavily capitalized tech giants.
How it boosts IPO prices
This is the part that deserves the most scrutiny, because the logic is not obvious.
Michael Burry argued that portraying AI as powerful enough to pose an extinction-level threat adds to the "hype and puffery" surrounding Anthropic and OpenAI ahead of their planned IPOs. The argument is that framing your product as world-historically powerful, even dangerously so, inflates its perceived importance and therefore its valuation. Safety reduces perceived regulatory and reputational risk for investors, making Anthropic attractive despite smaller consumer scale.
There is also a timing dimension. OpenAI has ruled out completing an initial public offering in 2026, with CEO Sam Altman saying concerns over AI safety make this an "ill-advised moment" to go public. Critics read that as a deliberate pause to let valuations stabilize before a cleaner public debut. The argument is: pause enough to IPO cleanly, then raise the ladder behind you.
How it slows competitors
The ongoing claim is that the industry's calls for regulation are actually a strategy of regulatory capture: powerful companies that already enjoy prominence may use regulatory stratagems to ice out or disadvantage smaller, less-resourced companies, thereby stifling competition.
A D.A. Davidson analyst said he thinks Anthropic and OpenAI are engaging in "monopolistic behavior." OpenAI has reportedly asked members of Congress for guidance about whether a coordinated, industrywide slowdown would violate antitrust law, according to Wired. The same analyst called it a "ladder pull," meaning you climb up and then pull the ladder up after you.
By warning lawmakers about the existential dangers of AI, the biggest frontier labs hoped to inspire expensive safety regulations that smaller startup competitors could never afford. By outlawing or heavily restricting open-source models, which allow small businesses and individual entrepreneurs to run cheap computing tools locally, major labs can eliminate low-cost competition.
The honest read
None of this requires cynicism as the only explanation. None of this requires the safety warnings to be insincere. Amodei, Altman, and their employees may genuinely fear their technology while favoring a regulatory structure that protects their commercial position. The question is what the proposed rules accomplish, including whom they prevent from competing.
The critics are not saying the danger is fake. They are saying that even if the danger is real, the proposed remedy happens to be extremely convenient for the people proposing it. That is a distinction worth holding onto, because it changes what you are actually arguing about.
How does regulation limit open-source AI
Regulation targets open-source AI in three specific ways: pre-release safety certification that only closed companies can practically meet, compliance costs that price out independent developers, and the structural impossibility of pulling a model back once the weights are public.
The core tension
The pro-regulation case from Anthropic, OpenAI, and Google DeepMind rests on a real problem with open-source models that closed models do not share. When a closed model turns out to be dangerous, the company can cut off access. When an open-weight model turns out to be dangerous, there is no switch to flip. As one governance paper puts it, there is "no single point of access that could be denied" because other actors may have already reproduced the model, weights and all. That is a genuine asymmetry, and regulators have noticed it.
How the rules would land in practice
The three mechanisms are distinct.
Pre-release certification. The regulatory proposals from Amodei and Hassabis require safety evaluations before a model reaches the public. For a closed company with a legal team, a safety division, and a government relations budget, that is friction. For an independent researcher or a small lab releasing weights on Hugging Face, it is a wall. The EU AI Act already demonstrates this pattern: it exempts open-source models from some requirements but carves out an exception for general-purpose AI, meaning the most capable open models lose the exemption precisely when it would matter most.
Compliance costs. Pre-release audits, documentation requirements, incident reporting, and certified alignment testing all have price tags. Fixed regulatory costs fall on everyone equally in dollars and on no one equally in pain. A lab with a billion dollars in funding absorbs them; a solo researcher or a five-person team does not. The result is not a safer ecosystem but a more concentrated one.
The non-decommissionability problem. Regulators argue that once model weights are released or leaked, the developer loses control and the model cannot be recalled. This is the structural argument for holding open-source to a higher pre-release standard than closed models, because post-release remedies simply do not exist. Critics, including analysts at the R Street Institute, counter that transparency is not the same as danger, and that open weights actually allow for deeper safety auditing than closed black-box systems.
Who benefits from that outcome
This is where the earlier conversation connects. If the regulatory framework is built around pre-release certification and compliance infrastructure, it consolidates the frontier among the companies already at the frontier. Meta and Nvidia have staked out the opposite position partly because Meta's Llama family is open-weight and any regulation written around the non-decommissionability argument hits Meta harder than it hits Anthropic. The critic case is that the safety framing, whatever its genuine merits, produces a rulebook whose practical effect is to slow the one development path that smaller players and independent researchers can actually use.
The honest answer is that both sides have something real. Open-source models do create genuine control problems once weights are public. Compliance costs do systematically favor incumbents. The question is whether those two facts justify the same regulation, and that is a question the people writing the rules have not yet settled cleanly.
Comments