OpenAI, Anthropic, and Google DeepMind CEOs Agree: AI Needs Urgent Regulation

The three CEOs pushing hardest in the race for ever more powerful artificial intelligence have, for the first time, publicly and simultaneously called for urgent regulation of the most advanced models. Sam Altman of OpenAI, Dario Amodei of Anthropic, and Demis Hassabis of Google DeepMind each published separate written proposals for AI oversight over the past several weeks, as noted by Axios in its "Behind the Curtain" analysis.

Three Different Prescriptions

Despite sharing the diagnosis that the most powerful models require oversight, the three leaders propose entirely different mechanisms. Amodei wants an agency with the power to block a model's release from day one if it fails independent safety testing. In his essay, he compares frontier AI models to airplanes, which must be certified before they can take to the air.

Frontier AI models, like airplanes, should be required to go through technical testing and auditing, and their release should be blocked or reversed as a threat to public safety if they do not meet high standards of safety - Dario Amodei, CEO of Anthropic

Anthropic's proposal holds that models exceeding a set compute threshold would undergo mandatory third-party testing in four areas: cybersecurity, bioweapons risk, loss of control over the system, and the model's ability to automate its own research and development. The government could block or reverse a model's release if the assessment finds unacceptable risk, with testing carried out either by an FAA-style government agency or by authorized private entities.

The FINRA Model and the IAEA Model

Hassabis, who previously outlined his own oversight concept, proposes a gentler starting point: an organization funded by the industry but federally supervised, modeled on FINRA, the US securities regulator. In his version, voluntary pre-release reviews could eventually evolve into mandatory market-entry rules.

Altman is taking a completely different approach, proposing an international solution. In a column for the Financial Times published in early July 2026, he described a US-led forum that would certify countries, companies, and safety standards, using access to the most powerful models and markets as leverage to enforce compliance. He pointed to aviation, global financial standards, and the International Atomic Energy Agency as models to follow.

AI will reshape the material conditions of human life on a scale that no technology has accomplished since the harnessing of electricity, and perhaps beyond even that - Sam Altman, CEO of OpenAI

Accusations of Locking In the Market

All three proposals share one criticism from skeptics: each, to varying degrees, favors companies that already have the resources to meet the requirements. OpenAI, Google, and Anthropic have teams of lawyers, safety staff, and established government relationships, which makes it far easier for them to navigate a complex certification process. Startups and open-source model developers, who today keep the market disciplined on price and accessibility, would face a much harder path to meeting the same standards.

The nature of the technology itself deepens the problem. Unlike airplanes or nuclear facilities, which inspectors can physically examine, AI models are trained inside data centers with very limited outside access. That makes verifying whether a given lab is actually following the agreed rules, or quietly developing technology while bypassing procedures, far harder than in the industries the three CEOs' proposals are modeled on.

What It Means for Poland and Europe

For the European, including Polish, AI market, the convergence of views among the three largest labs carries practical weight. The European Union has been rolling out its own AI Act for months, and starting August 2, 2026, it will begin enforcing rules on developers of the most powerful models. If the United States does end up creating a new federal or international oversight structure, European regulators will need to decide whether to pursue harmonized standards or maintain their own separate certification path.

For Polish companies using OpenAI, Anthropic, or Google models, this potentially means another layer of compliance requirements in the coming years, on top of the EU rules already in force. At the same time, Altman's proposal, based on certifying countries and companies with access to the most powerful models as a reward for compliance, could in practice affect which states and markets get priority access to the newest technology.

None of the three proposals currently exists as legislation or a binding international agreement. For now, they remain individual company positions, published as essays and opinion columns rather than draft rules adopted by Congress, the US administration, or any international body. Whether the shared diagnosis of three rivals translates into actual law now depends on the response of the White House and Congress, which have so far approached AI regulation cautiously and piecemeal.