OpenAI, Anthropic, and Google Plan a Self-Regulatory AI Standards Body

The three frontier labs are building their own regulator

OpenAI, Anthropic, and Google DeepMind are moving forward with plans for a joint industry-led AI safety standards body, tentatively named the Standards Authority for Frontier AI, or SAFA. The Information first reported the plan, and the three labs are pushing forward independently of government, citing a regulatory vacuum as Washington's AI stance stays voluntarist. The model is FINRA: an industry-run authority that defines safety commitments and supports third-party testing of frontier models, including testing the labs' own systems directly.

The initiative grew out of a proposal by Google DeepMind chief Demis Hassabis for a self-regulatory body, and it comes with a targeted launch by late 2026 or early 2027. The three labs have been holding working group meetings since July on how such an authority would operate.

Why the labs are doing this without Washington

The stated reason is a vacuum. The administration has dismissed safety concerns, declined to safety-test open-weight models under its voluntary framework, and pushed to keep pace with China. A self-run authority lets the labs set common testing benchmarks and disclosure practices without waiting for legislation that is not coming in this Congress.

The context makes the motive legible. This summer's agent incidents, the Hugging Face breach, the misalignment disclosures, and a Senate document demand all landed within weeks of each other, and the industry's own chiefs responded by asking for permission to slow down. A standards body is the institutional answer to a question the labs have started asking themselves: who verifies the verifiers, if the government will not?

The arrival calendar is crowded, and pointed

SAFA's targeted launch window lands amid a cascade of governance events: the first dedicated US-China AI safety dialogue with its proposed incident notification mechanism, a UN scientific panel warning that traditional safeguarding is unraveling, a bipartisan House push for pre-deployment testing requirements, and a royal summit in Scotland. Every institution has now weighed in, and each proposal differs: government notification channels, UN guardrails, statutory testing mandates, and now industry self-certification.

The overlap matters less than the incentives. A FINRA-style body funded and governed by the labs it oversees has a documented capture problem; FINRA itself operates under SEC oversight, while SAFA's backstop is the reputational kind. The open question is whether third-party testing under lab governance carries the independence that the phrase implies.

What readers should take from it

For anyone deploying frontier models, SAFA's eventual benchmarks would be the closest thing to a common safety baseline the industry has produced, and worth tracking for procurement decisions. For everyone else, the signal is simpler: the labs are acknowledging, by building this, that current safeguards are not the finish line. Whether self-regulation is a bridge to real oversight or a substitute for it is the decade's open question.