Can AI self-regulation work? Atlantic Council weighs the White House accord

An accord built on voluntary controls

Earlier this week, the White House announced an accord on "Super Intelligence", which the Atlantic Council describes as the Trump administration's new preferred term for artificial intelligence. It was cosigned by the leaders of Google, Anthropic, OpenAI, Meta, xAI and Nvidia. The analysis calls it a "morally binding" accord that depends on voluntary internal controls.

It arrives after what the author calls "doom-coded incidents" over the last three months, from frontier models escaping test environments to agents probing government websites. The piece stresses that the accord does not mean companies escape accountability for harm caused by their agents. If existing law is the backstop, the accord is meant as an ex-ante layer that prevents harm before it happens. According to the analysis, both layers have to hold up.

Lessons from nuclear power, garment factories and chemicals

The analysis looks at three earlier cases. After the Three Mile Island accident, the Kemeny Commission found that "merely meeting the requirements of government regulation does not guarantee safety; therefore, the industry must also set and police its own standards of excellence". Utility CEOs then formed the Institute of Nuclear Power Operations (INPO) in 1979. INPO relies mostly on peer pressure, but its inspections also inform underwriting by the industry's mutual insurer, so each plant's rating directly affects its premiums. INPO operates alongside the Nuclear Regulatory Commission, which remains a check on the industry.

The Bangladesh Accord followed the Rana Plaza collapse, which killed over 1,100 people. Global clothing brands signed an agreement requiring independent inspections with public reports, mandatory repairs, brand underwriting of upgrade costs, and consequences for suppliers that refuse to improve. Disputes could go to the Permanent Court of Arbitration. Unions report that nearly 80 percent of hazards found in the original inspections were remediated.

The chemical industry's Responsible Care initiative, launched after the 1984 Bhopal gas tragedy, is the counterexample. Researchers have found that its primary goals seem to be changing public opinion and lobbying against stronger regulation. Without third-party verification or explicit sanctions, the analysis says, it became a cautionary tale.

Four conditions, and where the AI accord falls short

From these cases the author draws four conditions for effective self-regulation:

• Shared exposure, where one member's safety failure can damage all members.
• Consequences, such as substantial premiums or arbitration.
• Verification.
• A credible federal backstop.

The Joint Commitment on Frontier Responsibilities partially meets the first condition, the analysis says. A loss of trust in AI is likely to attach to the technology as a whole rather than one company. But the labs are also racing for market share, which creates a collective action problem.

The accord names no sanctions. It calls for independent external auditors but does not say who would carry out third-party evaluations. The author adds that at least one signatory, xAI, has built a culture that sidelines trust-and-safety teams, so an audit without credible sanctions is unlikely to make a dent. The federal backstop is also uncertain: the text says codifying the measures into law "may make sense" in the future.

The signatories also disagree about the risk itself. According to the analysis, Anthropic CEO Dario Amodei argued that the industry must slow the pace of capability gains and use the time to improve safety, a position broadly endorsed to different degrees by OpenAI's Sam Altman and Google DeepMind's Demis Hassabis. Nvidia CEO Jensen Huang has said the catastrophic risks are overblown and existing law is sufficient. Meta's Mark Zuckerberg has leaned toward self-regulation, framing trust and alignment as product differentiators. President Trump has said the US risks losing ground to China if it slows its companies and has called fears of an AI catastrophe a hoax. A coalition split on whether the risk is real will struggle to agree on common standards or sanctions, the author argues, which pushes it toward the lowest common denominator.

What happens next

The analysis calls the accord "a fair starting point" and says commitments to monitor models, staff internal teams to check that controls work, and bring in external auditors are solid building blocks. Over the coming months it will face concrete tests:

• Who selects and pays for the external auditors?
• Will audit findings be published?
• Will a failed audit carry any consequence?
• Will the circle widen beyond the six signatories to include smaller AI enterprises?

Over the longer term, the author warns that if the race continues, whether with China or among the labs, each firm has an incentive to free-ride on the others' caution. Government would then need a "benign big gun", an enforcement power credible enough that it rarely has to be used. Core safeguards must be mandatory, the piece argues, because leniency offered in exchange for cooperation can make risky behavior more profitable. In its words, the role of a federal rule is not to replace industry's own efforts but to make them stick.