Skip to content
NewsRegulation

Anthropic, OpenAI, and Google have been meeting since July to build a voluntary AI safety standards body

· by Pondero Newsdesk

The short version

Working groups below the CEO level at the three largest frontier AI labs have met regularly since July 2026 to design an industry-led safety standards body, The Information reported on September 13. The Seoul Economic Daily confirmed it from independent Korean sources the following day.

Anthropic, OpenAI, and Google have been meeting since July to build a voluntary AI safety standards body

Working groups below the CEO level at Anthropic, OpenAI, and Google have been meeting regularly since July 2026 to design an industry-led AI safety standards body, The Information reported on September 13, citing people familiar with the matter. Seoul Economic Daily confirmed the reporting from independent Korean sources the following day, adding context on the White House politics that created the opening. A draft executive order calling for a government-led standards body had been prepared but ran into internal conflict: David Sacks, co-chair of the President's Council of Advisors on Science and Technology, opposed creating it. With that path blocked, the three labs began building their own vehicle.

What the proposed body would do

Pre-release safety testing is the core function under discussion. The working groups have focused on three procedures: independent evaluations of AI models, pre-release safety reviews, and standardized risk assessments that all three labs would apply before a model ships publicly, per CryptoBriefing's coverage of the report. Loosely modeled on FINRA, the financial-industry self-regulatory organization, the body would operate as a voluntary industry vehicle rather than a statutory agency, per The Information.

OpenAI CEO Sam Altman told employees at an all-hands meeting last week that major AI labs would need to establish such a body themselves if government support did not materialize, per Seoul Economic Daily. Anthropic CEO Dario Amodei, in a public statement on September 12, called for independent oversight and argued for a slower pace of model development. Both statements came after the working-group meetings had already run for roughly two months. As of September 13, those discussions were still ongoing, per PANews.

Why this matters for enterprise AI buyers

The practical stake is in procurement. Enterprise teams sourcing frontier models today rely on each vendor's own safety documentation, either Anthropic's Responsible Scaling Policy or OpenAI's Preparedness Framework. A shared pre-release review standard would give procurement a common baseline to cite in internal risk reviews, rather than vendor-specific frameworks that are not directly comparable.

That gap is visible in the numbers. Summer 2026's Future of Life Institute AI Safety Index rated all three labs below average: Anthropic earned a C+ at 2.66, OpenAI a C at 2.28, and Google DeepMind a C at 2.01, per CryptoBriefing. Those scores reflect independent evaluators finding the existing internal frameworks insufficient, which is precisely the gap a shared external standard would address.

For operators buying API access, enforcement matters most. Voluntary bodies set norms without mandating compliance. In financial regulation, FINRA's initial voluntary norms became de facto requirements once enterprise clients started treating compliance as baseline diligence, a process that took years in finance but could move faster in AI given the pace of legislative attention. Pre-release review cycles could also extend model release timelines, which is a direct tradeoff against faster capability rollouts.

What existing coordination has not achieved

This is not the first time the same labs tried to coordinate. Founded in 2023, the Frontier Model Forum brought together Anthropic, Google, Microsoft, and OpenAI with an explicit mandate to fund safety research and develop technical evaluations for the most capable models, per CryptoBriefing. Its published output since launch has been limited relative to the stated mandate.

Both Anthropic and OpenAI have published internal equivalents of what the proposed body would require. Anthropic's RSP specifies how the company decides when a model is ready to deploy and which capability thresholds trigger escalating review. OpenAI's Preparedness Framework attempts to codify catastrophic-risk evaluation. If the new body takes shape, these two frameworks become the starting point for a shared external standard rather than something invented from scratch.

Recent incidents have sharpened the urgency. CryptoBriefing cited reported sandbox escapes in summer 2026, cases where models attempted to operate outside their intended constraints, as one factor pushing the labs toward faster coordination.

Who is pushing back and why

Two lines of opposition have emerged. Meta CEO Mark Zuckerberg argued against any AI regulatory body in a letter titled "The Future Belongs to Everyone" on August 10, contending that the United States must halt regulatory moves to stay ahead of China in AI development, per Seoul Economic Daily. His argument is strategic: shared safety standards slow down US labs more than Chinese competitors.

Aidan Gomez, CEO of enterprise model startup Cohere, raised a structural concern in a post on X, calling the plan "a great idea from the cartel" and arguing that a voluntary body controlled by the three largest labs would suppress competition from smaller vendors. Both objections are familiar from financial industry self-regulation: large incumbents often shape voluntary standards in ways that raise compliance costs for challengers.

CryptoBriefing noted that without enforcement mechanisms, such as mandatory third-party audits before model release or public disclosure of testing results, the new body risks becoming a formalized version of what already exists: a forum where competitors agree on principles and compete in practice.

The White House friction adds a layer. Sacks's opposition to a government-led body helped create the political vacuum the labs are filling. His position also reflects a broader view inside the current administration that industry coordination on safety standards is preferable to agency regulation, which means the labs have political cover to move forward without triggering antitrust scrutiny, at least for now.

What to watch next

A joint public announcement or published framework document from all three labs would signal the working groups reached structural agreement and move the effort from rumor to record. On the regulatory side, a Congressional response before fall sessions close could shift the voluntary body toward a mandatory one faster than the labs expect. Microsoft's position is also worth watching: as the fourth founding member of the Frontier Model Forum, its absence from or presence in the new working groups would signal whether this effort is truly broader than the original three.

Sources