OpenAI, Anthropic and Google explore AI safety body as regulatory debate intensifies

OpenAI, Anthropic and Google DeepMind are discussing the creation of an industry-wide body to establish safety standards for advanced artificial intelligence models, as concerns over the pace of AI development collide with political resistance to tighter regulation in the United States.

The initiative reflects a growing effort among some of the industry's largest companies to develop common safeguards as AI systems become increasingly capable. But the proposal also raises questions about who should set the rules, how independent such a body could be and whether cooperation between leading competitors could create competition-law concerns.

The discussions follow a proposal by Google DeepMind co-founder Demis Hassabis for a U.S.-led organization modeled partly on the Financial Industry Regulatory Authority, or FINRA, the self-regulatory organization overseeing U.S. broker-dealers under the supervision of the Securities and Exchange Commission.

A safety body modeled on financial self-regulation

Hassabis outlined the concept in July, arguing for an organization that could establish technical standards and assess frontier AI systems before they reach the market. The proposed structure would be financed by industry participants but would also include independent experts and representatives of the open-source AI community.

The idea has since gained momentum. OpenAI confirmed this week that it has been working with Anthropic and Google DeepMind on AI safety for several weeks. The companies' discussions have focused on coordinating safety efforts without requiring an antitrust waiver, according to reporting cited by Reuters.

Under the proposed model, the most advanced AI developers could be expected to submit new systems for independent or industry-wide evaluations, particularly where questions of safety and potential misuse arise.

The comparison with FINRA, however, has an important legal limitation. FINRA operates within a framework established under U.S. law and remains subject to SEC oversight. The proposed AI organization would therefore need a legal foundation if it were to exercise authority beyond voluntary industry standards.

The competition-law question

The prospect of major AI companies jointly determining safety requirements has immediately raised concerns about competition.

A body dominated by the industry's largest developers could potentially establish standards that smaller competitors would find difficult to meet. Critics have also questioned whether cooperation between companies that compete directly in the frontier-AI market could inadvertently reduce competition or reinforce the position of incumbents.

The debate is particularly sensitive because the companies involved control some of the industry's most advanced models and have invested heavily in infrastructure, research and commercialization.

Aidan Gomez, chief executive of Canadian AI company Cohere, has argued that AI rules should not be written by a small group of leading laboratories alone. The broader technology ecosystem, including smaller companies and other stakeholders, would need to be represented in any credible framework.

Political resistance complicates the proposal

The industry initiative is developing against a different policy backdrop in Washington.

President Donald Trump and members of his administration have emphasized maintaining the pace of U.S. AI development, with competition from China a central consideration. Trump has dismissed some warnings about catastrophic AI risks, while the administration has pursued a comparatively limited approach to mandatory controls.

The White House has nevertheless introduced a voluntary process for evaluating new AI models. The precise mechanism and methodology have not been fully disclosed publicly, according to reporting on the initiative.

That creates a significant gap between voluntary industry safeguards and a formal regulatory regime backed by federal authority.

At the same time, several technology leaders have recently called for greater caution. Anthropic chief executive Dario Amodei has urged the industry to slow the pace of frontier-AI development so that safety measures can catch up. OpenAI chief executive Sam Altman and Hassabis have expressed support for coordinated safety measures, while discussions about independent evaluators have gained prominence.

From voluntary standards to international rules

The debate is no longer limited to the question of whether companies should test their own models.

Hassabis has presented the proposed industry body as a possible first step toward broader international standards. That ambition reflects the increasingly global nature of AI development, but it also exposes the difficulties of creating common rules among countries pursuing different economic and strategic objectives.

Some policymakers and technology figures argue that voluntary commitments cannot provide sufficient guarantees when the potential consequences of advanced AI extend beyond individual companies.

Others point to the strategic competition between the United States and China as a reason to avoid measures that could slow American developers more than their international competitors.

The result is a regulatory dilemma: governments want greater control over potentially dangerous technologies, while companies and policymakers remain concerned that excessive restrictions could weaken technological competitiveness.

The debate over who should regulate AI

The proposed oversight body therefore sits between two competing approaches.

One would give industry participants a substantial role in defining and enforcing technical standards, relying on their expertise and ability to evaluate rapidly changing AI systems. The other would place primary responsibility with public authorities, whose legitimacy derives from laws and democratic institutions rather than from participation in the technology market.

Senator Bernie Sanders has argued for binding international rules rather than relying solely on voluntary industry standards. His position reflects a broader argument that decisions involving potentially systemic AI risks should not be left exclusively to the companies developing the technology.

Meanwhile, critics of industry-led regulation have warned of what is commonly described as regulatory capture: a situation in which the companies being regulated gain excessive influence over the rules intended to constrain them.

The disagreement is likely to remain central as the United States considers how to balance AI innovation, national competitiveness, consumer protection and safety.

A test for the AI industry's willingness to cooperate

The proposed body faces a practical challenge as well as a political one. The companies discussing it remain fierce commercial competitors, with enormous financial incentives to develop increasingly capable systems ahead of rivals.

Any agreement to coordinate safety testing or slow development would therefore represent an unusual form of cooperation within an industry built around technological competition.

For now, the initiative remains at the discussion stage rather than constituting a new federal regulator. Whether it develops into a formal standards organization, receives government backing or remains a voluntary industry framework will depend in large part on the political and legal environment in Washington.

What is already changing, however, is the nature of the debate. The world's leading AI developers are increasingly discussing not only how quickly frontier systems can be built, but also who should determine the limits of that race.