OpenAI, Anthropic and Google DeepMind are working to create a body that would oversee risks linked to artificial intelligence, OpenAI told the Agence France-Presse news agency on Tuesday, September 15.
The goal would be to establish common safety standards as companies race to build increasingly powerful AI models.
A body modeled on financial regulation
The idea was proposed by Demis Hassabis, chief executive of Google DeepMind, who envisioned a structure inspired by the Financial Industry Regulatory Authority, known as FINRA, the American financial industry's self-regulatory body. Under the proposal, the new body would be funded by companies in the sector, with a board made up of both independent experts and representatives of open AI models, meaning those that are free to access, downloadable and modifiable.
Such an organization could evaluate the most advanced AI models and set common rules before they are deployed. Discussions between the three AI giants are still ongoing, according to OpenAI.
Google DeepMind, founded in 2010 and acquired by Google in 2014, has developed AI systems including AlphaGo and Gemini. OpenAI, the creator of ChatGPT, and Anthropic, founded by former OpenAI executives and known for its Claude models, are among its main rivals in building so-called frontier AI systems.
Growing concerns over AI safety
The proposal comes as calls to strengthen AI safety oversight multiply. Several recent incidents at OpenAI and Anthropic showed that models could behave unpredictably, breaking out of their confined testing environments to access platforms on the internet during tests.
Bilal Chughtai, a former Google DeepMind employee, recently said this kind of episode proved that AI could escape the supervision of its developers and carry out dangerous actions that could lead to a definitive loss of human control or the end of humanity.
In response to these risks, several industry leaders, including Dario Amodei of Anthropic, Sam Altman of OpenAI, Demis Hassabis of Google DeepMind and Elon Musk, have in recent days called for slowing the pace of AI development and putting in place shared safety standards.
A divided industry
The issue divides the sector. French company Mistral said some major players could use these calls for greater regulation to strengthen their market position against competitors.
Mistral said deploying an AI model safely requires suitable computing infrastructure, appropriate monitoring mechanisms and a strict separation of access rights.
Mistral AI is a Paris-based startup that has positioned itself as a European alternative to the dominant American AI labs, often championing open-weight models.
What regulators are doing
In early August, the Trump administration set up a process for evaluating new AI models, but on a voluntary basis. The European Union already has its own regulatory framework, the AI Act, adopted in 2024, which imposes different obligations on AI systems depending on the risks they pose.
