The three rivals each released a detailed summary of their views over the past five weeks – the same extraordinary period in which Washington twice intervened to limit or delay access to border models.
- We hear Meta’s Mark Zuckerberg is also working on his own memo.
Driving News: Hassabis’ proposal, released Tuesday, drew rare public praise in the highly competitive AI industry, including from Altman, Microsoft CEO Satya Nadella and even longtime rival Elon Musk.
- Anthropic co-founder Jack Clark called the framework “excellent” and wrote: “At this point, everyone at the frontier of AI agrees that third parties should test AI systems and use them to develop standards that inform policy.”
The Trump administration itself is torn: Publicly, it has championed deregulation and opposed anything resembling an “FDA for AI,” intent on not stifling U.S. innovation in the race with China.
- Privately, officials admit that a total hands-off approach is untenable: Cyber fears have already forced them to make improvised regulations twice this summer — first on Anthropic’s Fable and Mythos models, then on OpenAI’s GPT-5.6.
The big picture: Amodei, Altman and Hassabis (plus countless CEOs and investors) generally agree on a rough regulatory framework.
- Independent testing: All three want boundary models to be subject to external review before they are made public – a break with the industry’s old self-disclosure standard.
- A system of government: All three rely on outdated regulatory models and propose bodies that can set standards, certify compliance and limit access to border systems deemed too dangerous.
- America First: All three want the United States — not a fragmented patchwork of states or rival national regimes — to set the terms of a body with international reach.
- Threat Awareness: All three cite looming national security vulnerabilities, including dangerous cyber and biological weapons capabilities.
- Innovation protection: None of them call for comprehensive action against AI. The common target is the small class of boundary models that are powerful enough to generate catastrophic or strategic risks.
Where they disagree: The AI sponsors There are differing opinions as to whether the government itself should be the sole final arbiter.
- Amodei wants an FAA for AI: a federal agency with the power to immediately block the release of a model from day one.
- Hassabis wants a FINRA for AI: an industry-funded, federally supervised standards body that starts with voluntary pre-reviews and could evolve into mandatory market access rules.
- Altman, Writing in the Financial Times ($)pushes for an IAEA for AI: a US-led international forum that certifies countries, companies and security standards, using access to border models and markets as leverage for compliance.
Between the lines: OpenAI, Google and Anthropic already have the lawyers, security teams, government relations and technical staff to navigate a complex certification process. Startups and open source developers would face a much steeper climb.
- Critics fear that this could lead to regulatory co-optation: rules designed to make AI safer could lead to the largest AI companies establishing themselves.
The conclusion: The Wild West era of AI development is officially over. The people with the most money, the most computing power, and the most to lose from AI slowdowns are the ones most committed to regulation.
- Axios’ Zachary Basu contributed reporting.
Read the manifestos: Demis Hassabis…Sam Altman (April preview)…Dario Amodei.
https://www.axios.com/2026/07/16/ai-regulations-openai-anthropic-google
