AI Standards Body: Top Labs Plan to Regulate Themselves
3 min readThree of the biggest names in artificial intelligence are building the regulator they would rather answer to. Google, OpenAI and Anthropic are moving ahead with plans for an independent organization that would set safety standards for frontier AI models, tentatively named the Standards Authority for Frontier AI, or SAFA.
Why an AI Standards Body, and Why Now
Frontier AI has been governed mostly by promises. Since 2023 the leading labs have published voluntary safety commitments, model cards and internal risk frameworks, each written to its own template and enforced by nobody outside the company. Governments have moved unevenly. The European Union has binding rules, several US states have passed their own, and federal legislation has repeatedly stalled.
That gap leaves an awkward situation. The companies with the most capable models are also the ones with the deepest expertise to evaluate them, and no neutral party holds the results.
What the Three Labs Are Proposing
SAFA would operate independently of its founders and focus on four jobs, according to BankInfoSecurity: supporting third-party organizations that test models before deployment, defining how developers report safety and security incidents, turning the labs’ voluntary commitments into something measurable, and setting qualifications for independent auditors of both models and the companies building them.
The effort is further along than a concept. As PYMNTS reports, the three companies have approached prominent figures about roles in the group, including former White House AI policy adviser Sriram Krishnan, former Biden administration technology official Arati Prabhakar, former Secretary of State Condoleezza Rice and venture capitalist David Friedberg. A launch is targeted for late this year or early 2027.
Why It Matters
An industry writing its own standards is not new. Financial exchanges, accounting firms and the early internet all built self-regulatory bodies, and the results ranged from genuinely useful to purely cosmetic. What separates the two is usually whether the body can say no and make it stick.
SAFA’s credibility will rest on that question. If it can set an audit qualification its own founders fail, or publish an incident report they would rather bury, it becomes real infrastructure. If it cannot, it becomes a well-staffed argument against government rules. Watch who accepts the senior roles, and whether any lab outside the founding three signs on.
Regulators in Washington and Brussels are still deciding how far to go with frontier models. The three labs building those models have decided not to wait for them.