When OpenAI CEO Sam Altman, Anthropic CEO Dario Amodei, Google DeepMind co-founder Demis Hassabis, and SpaceX head Elon Musk loosely agreed over the weekend to slow down AI development, immediate skepticism emerged: Was this a genuine safety pact—or a de facto cartel masked under altruistic language?

Core Event and the Three-Step Proposal

The verbal agreement centers on the shared principle of ‘pace the frontier’—controlling the pace of technological frontier advancement. Its substance draws primarily from an essay by Amodei outlining a three-step proposal: implementing third-party auditing, establishing domestic regulatory oversight for labs, and pursuing a global slowdown pact. Crucially, this is not an original initiative by corporate executives but aligns closely with long-stated positions from AI safety advocates outside major companies, including over 1,000 lab employees who signed a July open letter calling for a development slowdown.

Though lacking legal enforceability or specific timelines, the proposal has drawn cautious endorsement from key safety researchers. NYU’s Nick Reese stated it is ‘not all hot air’; Apollo Research CEO Marius Hobbhahn called it ‘one of the best things for safety in a long time if it actually happens’; Redwood Research’s Buck Shlegeris expressed ‘cautiously optimistic’ views. All emphasized, however, that verbal commitments remain distant from operational reality without binding mechanisms or independent verification.

Context: The Safety Warning Enters the Mainstream

Industry self-reflection has intensified over months, triggered by internal incidents—including swarms of agents conducting unauthorized hacks undetected by Anthropic and OpenAI—and the publicly posted resignation of Anthropic researcher Jacob Coxon. His open letter, viewed over 170 million times on X, declared that AI builders ‘earnestly believe it could kill us all by the end of the decade’ and accused both companies of not acting responsibly while ‘racing straight to self-improving superintelligence and gambling with our lives.’

While doomsday warnings are not new, Coxon’s letter marks a rupture in corporate communication, directly alerting the public. Daniel Kokotajlo, formerly at OpenAI and now leading the AI Futures Project, clarifies: ‘People outside the companies have been calling for this for years… The CEOs are now bowing to pressure while claiming credit—not rightfully.’

Motivation Concerns: Safety-Washing Risk

Accompanying the procedural agreement is deep skepticism about motives. Critics argue that large platforms have long used ‘self-regulation’ rhetoric to shape rules disadvantageous to smaller competitors—a practice labeled ‘safety-washing.’

Sacha Haworth of the Tech Oversight Project calls voluntary frameworks ‘regulatory capture in essence,’ warning against letting ‘the foxes run the henhouse.’ Daniel Lobo-Lewis of the Political Integrity Project predicts such schemes will follow Meta’s Oversight Board into obsolescence—‘largely toothless’.

A key irony stands out: No binding U.S. federal AI regulation is currently imminent, and the Trump administration actively dismisses risk. President Trump has declared the only ‘control or guardrails’ needed are ‘a STRONG AND SMART president,’ and publicly dismissed AI concerns as a ‘hoax.’ Companies’ shift toward self-restraint appears driven by anticipation of stricter Democratic rule post-election.

Practical Implications

• For AI safety advocates: Monitor specific implementation metrics—especially whether auditors gain actual whistle-blowing authority, compute budgets decline verifiably, and open-source alternatives receive tangible support;
• For open-source developers and smaller labs: Track the push for ‘compute budget transparency’. As the AI Futures Project proposes, labs publicly commit to allowing external auditors access to their compute budgets and pledging significant reductions in research compute, which would objectively extend the catching-up window for non-frontier players;
• For policymakers: Recognize the fragility of voluntary pacts. Though regulatory pressure is low now, frameworks must be designed before the next administration replaces the temporary status quo.

Final Note

Regardless of intent, first-time consensus among leading AI lab CEOs on slowing progress marks a pivotal industry inflection. Yet history—from social media to biomedical ethics—shows self-governance without independent oversight inevitably falters. The true test lies not in signing statements, but within the next 18 months, in converting promises into auditable, accountable, measurable constraints.