The world's three leading AI laboratories are drawing up plans for an industry-run body to set safety standards for the most powerful artificial intelligence models, according to reporting by The Information published this week.
The group, provisionally called the Standards Authority for Frontier AI (SAFA), would act as an independent organisation that tests advanced models before release, supports independent evaluations, and sets common benchmarks for how developers report safety and security incidents.
The coalition could launch by the end of 2026 or early in 2027.
The move follows months of debate over how frontier AI should be overseen. Earlier this month, Anthropic chief executive Dario Amodei published an essay calling for the industry to slow down the release of frontier models so that safety mechanisms can catch up.
Google DeepMind co-founder Demis Hassabis had already proposed in July that the sector create a self-regulatory standards body modelled on the Financial Industry Regulatory Authority, the industry-funded organisation that polices Wall Street brokerages.
What the Watchdog Would Do
SAFA is expected to develop approaches to pre-deployment testing, incident reporting standards, and qualifications for the auditors who assess advanced systems.
The three companies are still debating the exact scope of the body, including whether it should conduct model evaluations itself or leave that work to outside laboratories.
That testing role currently falls to the US Center for AI Standards and Innovation, a Commerce Department unit. People familiar with the discussions say some in the group believe the centre lacks the technical and financial resources needed for thorough inspection of complex frontier systems.
The organisers have approached several high-profile figures about leadership roles, including former White House AI policy adviser Sriram Krishnan and former White House science director Arati Prabhakar for the chief executive position.
Former US Secretary of State Condoleezza Rice is reportedly under consideration for board chair.
Criticism and Regulatory Capture Concerns
The plan has already drawn criticism. Opponents accuse the three laboratories of attempting regulatory capture, warning that standards written by the companies being regulated could raise barriers for newcomers trying to build cutting-edge models.
Nvidia chief executive Jensen Huang and Meta chief executive Mark Zuckerberg have both publicly opposed the approach.
Critics argue that high testing and compliance costs could entrench the three leading labs and freeze out smaller developers, particularly open-source teams that lack the capital for specialised third-party audits.
It is also unclear whether other AI developers would be allowed to join the body as their models catch up to the frontier, and the legal authority behind any standards it issues has not been announced.
From Federal Oversight to Self-Regulation
The industry-led plan emerged after a federal approach stalled. A draft White House executive order that would have brought federal supervision of frontier AI failed to secure enough backing, prompting the three laboratories to press ahead on their own.
For enterprise users, a working SAFA could eventually provide a shared reference point for comparing vendors' safety claims before deploying advanced AI systems.
But with the body's testing powers, leadership, and enforcement mechanisms still undecided, analysts caution that it should be seen as a potential layer of oversight rather than a guarantee that frontier models will be independently certified.
