The main competitors in AI are coming together to discuss AI regulation.

      The leaders of the most advanced AI companies rarely find common ground. However, this week, three of them reached a consensus on a significant issue: the need for regulation of frontier AI, and it must happen soon.

      Over the past five weeks, the heads of Google DeepMind, OpenAI, and Anthropic each released a memo detailing how to govern the most capable AI models, as reported by Axios. Their assessments align closely with one another.

      Where they concur

      All three leaders advocate for independent, external testing of frontier models prior to their launch, which marks a departure from the industry’s previous reliance on self-reporting. They also seek a unified organization to establish standards, ensure compliance, and limit access to models deemed overly risky.

      They agree on leadership as well, favoring U.S. oversight rather than a mix of state or conflicting national regulations. They emphasize immediate national-security threats, including cyberattacks and bioweapons. None of them calls for widespread restrictions; instead, they focus on a small subset of the most powerful models.

      Where they differ

      However, they diverge on the extent of government intervention. Amodei, who has pushed for a coordinated halt on frontier AI, suggests creating an FAA-like federal agency that could outright block releases.

      Conversely, Hassabis proposes a FINRA-style organization that would be funded by the industry but under governmental supervision, starting with voluntary evaluations. Altman, writing in the Financial Times, proposes an IAEA-style international body that would leverage model and market access.

      An uncommon agreement

      Hassabis's proposal, published on Tuesday, received unexpected commendations amid a highly competitive arena. Altman described it as “thoughtful.” Microsoft’s Satya Nadella noted that the aim is to foster “a frontier ecosystem that encourages innovation and choice.” Even Elon Musk, who typically does not align with Altman, referred to it as “a good starting point.” Jack Clark of Anthropic praised the framework as “excellent.”

      The timing is significant. During the same five weeks, Washington intervened twice to restrict or postpone the release of frontier models, first Anthropic's Fable and Mythos, and then OpenAI’s GPT-5.6, citing cybersecurity concerns. The Trump administration has publicly opposed regulation, yet internally seems uncertain about the effectiveness of a hands-off strategy. It is also reported that Meta’s Mark Zuckerberg is crafting his own memo.

      The downside

      Not everyone is in favor. OpenAI, Google, and Anthropic have the legal resources, security teams, and connections in Washington to navigate a stringent certification process, but startups and open-source developers lack these advantages. Critics caution that this could lead to regulatory capture, where safety regulations inadvertently solidify the dominance of the largest labs. This is the same group of competitors that recently clashed at the G7 and is now strongly advocating for these regulations.

      Published July 17, 2026 - 10:46 am UTC