• Google (GOOG), OpenAI, and Anthropic are quietly negotiating a voluntary, industry-led AI safety standards body, though no formal organization has been announced.
  • The initiative, which remains at the working-group stage, aims to establish common model evaluations and pre-release review procedures.
  • OpenAI policy chief Chris Lehane confirmed the collaboration, arguing that an antitrust waiver should not be necessary for this kind of safety coordination.

A Voluntary Watchdog Takes Shape

Efforts to create a voluntary, industry-led AI safety standards body are gaining momentum, with Google, OpenAI, and Anthropic reportedly working together to establish common protocols for evaluating and releasing frontier AI models. The talks, which have been ongoing since July, stem from a proposal by Google DeepMind’s Demis Hassabis for a body modeled in part on the Financial Industry Regulatory Authority (FINRA), according to people familiar with the matter. OpenAI policy chief Chris Lehane publicly confirmed that the companies had worked together on safety for several weeks, adding that an antitrust waiver should not be necessary for this kind of collaboration.

The initiative, however, remains in its early stages. No final charter, governing board, funding plan, membership rules, or launch date has been established, and the group would not have legal authority to compel model developers to pause or alter releases. “It’s a working group, not a regulator,” one person close to the discussions said, emphasizing that the effort is meant to complement, not replace, government oversight.

Why Financial Scale Matters

The talks come amid a capital-intensive AI race that has seen unprecedented investment. Alphabet, Google’s parent company, reported second-quarter revenue of $119.8 billion, up 24% year over year, with Google Cloud revenue surging 82% to $24.8 billion. The company spent $44.9 billion in capital expenditures during the quarter and raised its full-year 2026 capex outlook to as much as $205 billion, underscoring the massive infrastructure demands of frontier AI development.

OpenAI, meanwhile, is reportedly considering financing at roughly a $1.5 trillion valuation, up from a prior $730 billion valuation, according to reports cited by the New York Times. The company has raised more than $180 billion since 2015. Anthropic recently announced a $65 billion Series H at a $965 billion post-money valuation, with third-party estimates placing its annualized revenue run rate at about $65 billion by July 2026.

These enormous sums create strong incentives to move models to market quickly. Common safety testing could increase confidence among enterprise customers and governments, but it could also add cost, delay, and disclosure friction for companies building or releasing advanced systems.

Independence Concerns Loom

The central controversy surrounding the proposed body is whether it can be independent enough to earn public trust. Critics worry that a body paid for or dominated by the leading labs could set weak rules, protect incumbents, or serve as a substitute for binding public regulation. The FINRA model has appeal because it creates a specialized oversight mechanism, but AI differs from financial intermediation: the largest labs both create the risks and would help set the tests, thresholds, disclosures, and sanctions.

“The key question is governance,” said a policy expert who spoke on condition of anonymity. “If the same companies writing the checks are also writing the rules, you have a credibility problem.”

A U.S.-centered industry body could become influential internationally only if its standards align with—or can be mapped to—the European Union’s AI Act, which imposes obligations on providers of general-purpose AI models. The Commission’s enforcement powers under this framework apply from August 2, 2026. If standards diverge, developers could face a fragmented compliance landscape.

A Familiar Playbook

This is not the first cross-company AI safety initiative. In July 2023, OpenAI, Google, Microsoft (MSFT), and Anthropic launched the Frontier Model Forum, committing to share practices, advance safety research, and cooperate with governments and civil society. The present discussion appears more consequential because it reportedly contemplates a structured standards body, shared model testing, pre-release review, and possible coordinated action when risks cross agreed thresholds.

Still, the current plan is unannounced and may not become an operational institution. The most important immediate signals will be whether the companies announce an independent governance structure, publish a shared testing methodology, include outside experts, and commit to incident disclosure.

What to Watch

Antitrust scrutiny will remain relevant. Competitors coordinating on technical safety standards can be pro-competitive, but regulators will watch closely for information sharing that could become coordinated product timing, access restrictions, or market exclusion.

In the near term, expect further negotiation over scope: cyber capability, biological-risk enablement, autonomy, model exfiltration, and deceptive behavior are likely candidates for common evaluation.

Three broad scenarios are plausible: a credible complement to regulation that raises baseline safety practice; an industry pledge with limited force that intensifies calls for legislation; or a contested global standard that competes with EU and Chinese governance rules. The proposal’s success will ultimately turn on four design choices: independent governance, measurable thresholds, real consequences for noncompliance, and enough public transparency for outsiders to assess whether safety claims match actual practice.

Representatives for Google, OpenAI, and Anthropic did not respond to requests for comment.

Correction: An earlier version of this article stated that the proposed body would have authority to compel model developers to pause releases. It would not.