Google, OpenAI, and Anthropic working to launch independent AI safety standards body

The three companies building the most powerful AI systems on the planet have decided they should probably agree on some ground rules. Google DeepMind, OpenAI, and Anthropic have been in working-group discussions since July 2026 to establish an independent, industry-led body focused on AI safety standards, with a target launch window of 2026 or 2027.

The goal: shared protocols for evaluating advanced frontier models, independent safety assessments, and pre-release checks before these systems reach the public.

What’s on the table

Anthropic CEO Dario Amodei laid out the intellectual case for the effort in a September 12, 2026, essay that called for external evaluators to play a central role in AI safety. His argument centers on the pace of AI development, which has consistently outrun the ability of any single company, or any government, to keep up with safety implications.

OpenAI CEO Sam Altman has echoed the sentiment, framing independent industry standards as a necessity given how slowly government regulation has moved. OpenAI’s global policy chief Chris Lehane confirmed on September 15 that coordination among the three firms has been a strong focus, and pushed back on the idea that their collaboration would require antitrust exemptions.

Lehane dismissed the notion that the companies would need antitrust waivers for their joint safety work, characterizing the effort as fundamentally about risk management rather than market coordination.

The credibility gap

There’s a notable tension in this initiative. All three companies scored C+ or lower on the 2026 Future of Life Institute AI Safety Index. The AI Safety Index evaluates companies on transparency, safety implementation, and responsible development practices.

This isn’t their first attempt at collective action, either. Google, OpenAI, and Anthropic co-founded the Frontier Model Forum back in 2023, a parallel initiative aimed at advancing AI safety research and establishing best practices. The new effort appears designed to go further, creating something with more structure, independence, and teeth than the Forum has provided.

Each company also maintains its own safety framework. Anthropic operates under what it calls a Responsible Scaling Policy, which ties the deployment of increasingly capable models to specific safety benchmarks. Google DeepMind and OpenAI have their own internal evaluation processes. The proposed standards body would presumably sit above these individual frameworks, creating shared benchmarks that any frontier model would need to clear.

Why it matters beyond AI

The lack of a finalized governance model or funding structure after months of talks raises real questions about execution. As of mid-September 2026, no formal organization exists. No leadership has been named. No charter has been published.

Smaller AI companies should be watching closely. If the three largest labs succeed in establishing industry-wide safety standards, those standards will inevitably reflect the capabilities and resources of companies operating at massive scale. Compliance costs that are rounding errors for Google could be existential for a 50-person AI startup, potentially accelerating consolidation in a market that already skews heavily toward a few dominant players.