AI’s fiercest rivals just agreed on one thing: regulate frontier AI now

The chiefs of Google DeepMind, OpenAI, and Anthropic have each published memos calling for independent testing and a single US-led regulatory body for frontier AI models, citing national-security risks. While they agree on the need for regulation, they differ on enforcement: Anthropic wants a federal agency with blocking power, Google DeepMind proposes an industry-funded body, and OpenAI suggests an international forum. Critics warn the rules could entrench the largest labs at the expense of startups.

The people building the most powerful AI rarely agree on much. This week, three of them agreed on one big thing. The frontier needs regulating, and soon. Over five weeks, the chiefs of Google DeepMind, OpenAI and Anthropic each published a memo on how to police the most capable AI models, Axios reported https://www.axios.com/2026/07/16/ai-regulations-openai-anthropic-google . Their diagnoses line up almost exactly. Where they agree All three want frontier models tested by independent outsiders before release. That breaks with the industry’s old habit of self-reporting. All three also want a single body to set standards, certify compliance and restrict access to models judged too dangerous. They agree on who leads, too. Each wants the US in charge, rather than a patchwork of states or rival national rules. Each points to near-term national-security risks, from cyberattacks to bioweapons. And none calls for a broad crackdown. The target is the small class of the most powerful models. Where they split The rivals part ways on how hard the government should hold the button. Amodei, who has also urged a coordinated pause https://thenextweb.com/news/anthropic-urges-a-coordinated-verifiable-pause-for-frontier-ai on frontier AI, wants an FAA for AI: a federal agency that can block a release outright. Hassabis wants a FINRA-style https://thenextweb.com/news/demis-hassabis-frontier-ai-standards-body-finra body, industry-funded but government-overseen, starting with voluntary reviews. Altman, writing in the Financial Times https://www.ft.com/content/0c2e1077-f658-4b3d-9040-602615c961ca , wants an IAEA-style international forum that uses access to models and markets as leverage. A rare truce Hassabis’s plan, published Tuesday, drew unusual praise across a bitterly competitive field. Altman called it “thoughtful.” Microsoft’s Satya Nadella said the goal was “a frontier ecosystem that promotes innovation and choice.” Even Elon Musk, no friend of Altman’s this week, called it “a good starting point.” Anthropic’s Jack Clark called the framework “excellent.” The timing is not random. Over the same five weeks, Washington twice stepped in to restrict or delay frontier models, first Anthropic’s Fable and Mythos https://thenextweb.com/news/anthropic-claude-mythos-warning , then OpenAI’s GPT-5.6, over cyber fears. The Trump administration is publicly anti-regulation, but privately less sure a hands-off approach can hold. Meta’s Mark Zuckerberg is said to be drafting a memo of his own. The catch Not everyone is cheering. OpenAI, Google and Anthropic already have the lawyers, security teams and Washington https://thenextweb.com/news/silicon-valley-ai-regulation-trump-biden-irony-framework ties to handle a heavy certification process. Startups and open-source developers do not. Critics warn the result could be regulatory capture: safety rules that quietly entrench the biggest labs. It is the same set of rivals who sparred at the G7 https://thenextweb.com/news/g7-ai-summit-altman-amodei-hassabis weeks ago, now lobbying hardest for the rules. Get the TNW newsletter Get the most important tech news in your inbox each week.