OpenAI, Google and Anthropic draft a standards body, SAFA

Google, OpenAI and Anthropic are getting closer to launching an independent body that would write safety standards for frontier AI, The Information reported. The working name is the Standards Authority for Frontier AI, or SAFA. In a September 27 summary of that report, Proactiveinvestors wrote that SAFA would operate independently and fill gaps in government oversight. President Donald Trump, meanwhile, opposes additional AI guardrails and says the US and China should leave AI development on its current course.

OpenAI and Anthropic are also reported to be working on a deal of their own. Each would get access to the other’s commercially available models for independent stress tests, looking for vulnerabilities and unexpected behaviour, and data obtained during those evaluations could not be freely retained. So far the reports describe discussions only.

The three companies have been here before. On July 26, 2023, together with Microsoft, they announced the Frontier Model Forum, which promised to work on “advancing technical evaluations and benchmarks” and a public library of solutions “to support industry best practices and standards.” The Forum still lists identifying best practices and supporting standards development as one of its three mandates, so the founders gave themselves SAFA’s stated job three years ago. Microsoft is not among the companies named in the SAFA reports.

How the coordination thread has moved this autumn:

• September 12: Dario Amodei publishes “We Must Pace the Frontier”, arguing for slower capability gains so safety work can keep up, and asking for a narrow government waiver so rival labs could coordinate on safety without running into antitrust law.
• September 15: OpenAI’s global policy chief Chris Lehane says the labs need no such waiver, per Bloomberg, and that OpenAI has been working with Anthropic and Google DeepMind on safety for several weeks.
• September 23: Sam Altman and Amodei address the UN Security Council on AI safety.
• September 27: the SAFA plan is reported.

At the Security Council, Altman said democratic institutions and governments should shape the major decisions about AI. Amodei called for international agreements and global standards for testing new models. Neither described SAFA as a replacement for government oversight. The plan as reported positions it as a stopgap for the gaps.

OpenAI has separately called on the US to lead international work on technical standards for advanced AI, and that includes systems capable of recursive self-improvement, where models help build the models that replace them.