Google, OpenAI, and Anthropic Plan to Form AI Safety Self-Regulatory Body
nashnova research
The three leading AI companies are building SAFA, an industry body to set frontier-AI safety standards — a critical attempt to write their own rulebook while U.S. government regulation remains stalled.
What is SAFA, and why now?
Google, OpenAI, and Anthropic are pushing to launch the Standards Authority for Frontier AI (SAFA), targeting late this year or early 2027.
The three originally envisioned a public-private body with the federal government, but the plan stalled inside the Trump White House — which told the companies to reach industry consensus first, then talk about government involvement.
This means → SAFA is now on an "operate independently first, seek government partnership later" track — essentially industry self-rescue in a policy vacuum.
What would this body actually do?
Four core functions: support third-party pre-deployment testing, standardize safety-incident reporting, turn each lab's prior voluntary commitments into concrete and auditable rules, and set qualification standards for independent auditors.
In plain terms = each company has said "we'll develop AI responsibly," but *how*, *who checks*, and *what gets checked* were never spelled out — SAFA aims to make those vague pledges enforceable and auditable.
The working group is also debating whether SAFA should run its own safety tests to fill gaps in CAISI — the Commerce Department's AI standards and testing center.
Who would lead it, and what does the shortlist signal?
CEO candidates include Sriram Krishnan, a former VC who served as a senior AI-policy adviser in the Trump administration, and Arati Prabhakar, who led the White House science office under Biden — one from each party.
Board-chair candidates include former Secretary of State Condoleezza Rice and entrepreneur David Friedberg; science-adviser candidates include METR founder Beth Barnes and OpenAI Foundation board member Paul Christiano.
This reflects an effort to secure three things at once: political endorsement, public credibility, and technical legitimacy. But some candidates say they have not been formally approached, and the list may still shift.
Why isn't the government handling this itself?
U.S. AI regulation is stuck on two fronts: Congress is gridlocked ahead of midterms, and the White House is cautious about pushing new AI policy.
Meta CEO Mark Zuckerberg and Nvidia CEO Jensen Huang have both publicly questioned tighter regulation — their opposition helped shelve an earlier White House executive-order draft.
This means → waiting for government action is not realistic in the near term. SAFA is the industry saying: rather than wait for a law that may never arrive, we'll build the framework ourselves.
What is the biggest open question?
SAFA is modeled on FINRA — the Financial Industry Regulatory Authority — a self-regulatory body built by the industry but with real enforcement power over its members.
In plain terms = FINRA can discipline brokers because you cannot do business without membership. The AI industry has no such licensing gate today, so what would compel SAFA members to actually follow the rules? That is the make-or-break question for the entire initiative.
The concept was first floated publicly by Google DeepMind CEO Demis Hassabis in a July article. The three companies have been advancing discussions since — but the road from proposal to operational body is still long.
市场有风险，内容仅供研究参考，不构成投资建议。