AI Giants May Slow Down Together as Safety Pact Debate Rages

On September 13, Cohere co-founder Aidan Gomez posted on X criticizing current proposals for third-party AI safety audit mechanisms, arguing they amount to "power projection" rather than real solutions. The post was widely shared and ignited a fierce debate in the developer community over self-regulation and open weights.

What's confirmed

• The mechanisms Gomez described include: full audits requiring "employee-level access" to a company's entire operations; the ability to shut a company down outright on safety grounds if it's deemed "not safe enough"; and an arrangement where China wouldn't comply, yet everyone else must submit or face chip supply cutoffs. He sarcastically called this mindset a "cartel."
• Gomez listed what he sees as today's real risk sources: models exhibiting "hack" behavior when OpenAI/Anthropic run hacker tasks on models in poorly sandboxed environments.
• Former OpenAI VP of Research Eric Mitchell clashed with Gomez from afar: critics questioned who appoints and funds verifiers, whose interests they represent, and worried this could concentrate AI development power in the hands of a few.
• Josh Purtell offered a counterargument: if a new lab achieves a breakthrough that confers an unfair advantage, verification rules would force it to hand over all its secrets to a set of third parties who would inevitably leak them—right into the hands of incumbents like OpenAI.

Why it matters

• The dispute cuts to the core tension in AI governance: supporters see external audits and independent verification as essential to keeping frontier models under control, while opponents argue the enforcement model could devolve into incumbents colluding to raise barriers, squeeze out competitors, and hand outsized discretion to a handful of verification bodies.
• The debate unfolds at a pivotal moment as regulatory frameworks take shape, and the divide between positions—open weights, self-regulation, mandatory third-party verification—will directly shape future policy.

2026-09-13 ~ 2026-09-13 · 25 related posts

Full story(17 episodes)→

• Episode 1: OpenAI employees speak out urging slower AI development over extinction risk(2026-09-10, 4 posts)
• Episode 2: OpenAI asks Congress whether coordinated AI slowdown would violate antitrust law(2026-09-11, 4 posts)
• Episode 3: Dario's "Pace the Frontier" Plea Draws Rare Backing from Altman and Musk(2026-09-11, 56 posts)
• Episode 4: Rival AI leaders Amodei, Altman and Musk align on slowing the frontier, sparking debate(2026-09-11, 62 posts)
• Episode 5: Anthropic CEO Calls for Slowing Down Frontier AI Development(2026-09-12, 128 posts)
• Episode 6: AI Audit Independence Questioned as METR's Ties to Labs Draw Fire(2026-09-12, 19 posts)
• Episode 7: AI Giants May Slow Down Together as Safety Pact Debate Rages(2026-09-13, 25 posts)
• Episode 8: The Open Weights Regulation Debate: Between Tiered Control and Full Lockdown(2026-09-13, 16 posts)
• Episode 9: OpenAI researcher: taming superintelligence is science, not pacing deals(2026-09-13, 3 posts)
• Episode 10: 'China Will Slow Too If We Do' Claim Mocked as AI Race Meme Spreads(2026-09-13, 4 posts)
• Episode 11: Open-source camp pushes back on frontier slowdown pledges(2026-09-13, 4 posts)
• Episode 12: OpenAI Researcher: Internal Progress Far Exceeds Public Perception(2026-09-13, 4 posts)
• Episode 13: Debate over AI labs' regulatory push: self-protection or regulatory capture(2026-09-13, 10 posts)
• Episode 14: Chollet Warns Frontier Lab Safety Proposals Risk Regulatory Capture(2026-09-13, 3 posts)
• Episode 15: Hassabis Backs Dario's AI Slowdown Call as Standards Bodies Weigh In(2026-09-13, 2 posts)
• Episode 16: Dario Amodei Says AI Is Still on an Exponential Curve with No Wall in Sight(2026-09-13, 2 posts)
• Episode 17: Rival AI CEOs Unusually Align in Calls to Slow Down AI(2026-09-13, 2 posts)

• Sam Altman hints at imminent pact among leading AI companies to slow development and address safety risks — jeremyakahn · 2026-09-13
• Dario's Call to Slow AI and Stop Open Source Draw Fire From Chamath and Critics — IgorCarron · 2026-09-13
• Neel Nanda: embedded evaluators are a first step, enforceable agreements needed — NeelNanda5 · 2026-09-13
• Neel Nanda: current AI progress unsafe without pacing, strike agreements now — NeelNanda5 · 2026-09-13
• Neel Nanda: embedded evaluators are a great first step, but we need enforceable agreements — NeelNanda5 · 2026-09-13
• DeepMind's Neel Nanda: Current AI Progress Is Clearly Unsafe Without Pacing Agreements — NeelNanda5 · 2026-09-13
• [source] Cohere Founder Criticizes AI Safety "Cartel" Proposals: Full Access or Shutdown, China Exempt — ns123abc · 2026-09-13
• Cohere CEO Aidan Gomez: third-party AI audits are power projection, not real fixes — yacineMTB · 2026-09-13
• Chamath accuses Dario of wanting to kill open source under guise of 'slowing AI' — markjeffrey · 2026-09-13
• Jason to Dario Amodei: slow Anthropic if you want, but don't lobby government against open source — aronchick · 2026-09-13
• Open Letter to Dario: If You Mean It, Mandate Open Weights for All Public AI Models — cephaloform · 2026-09-13
• [source] Eric Mitchell and Aidan Gomez clash over whether independent AI verifiers would concentrate power — ericmitchellai · 2026-09-13
• [source] Arguing third-party model audits would slow frontier labs the most — JoshPurtell · 2026-09-13
• willcb: Anthropic must pace releases for brand's sake, giving open source room to catch up — willcb · 2026-09-13
• Sam Altman says OpenAI, Anthropic and DeepMind may agree to slow down AI together — rohanpaul_ai · 2026-09-13
• Altman says top AI leaders meeting to slow down AI 'will happen' and may be underway — 4KTV · 2026-09-13
• Aidan Gomez slams AI safety 'cartel' proposals as devs argue self-regulation vs open weights — aronchick · 2026-09-13
• Rob LeClerc: 'pacing' slows public releases, not labs' internal development — and that gap is the real danger — robleclerc · 2026-09-13
• Dario's pacing essay gets scrutiny: safety concerns mixed with business interests, critics say — ziv_ravid · 2026-09-13
• Sam Altman endorses pacing the frontier and commits to independent evaluators with employee-like access — Turn_Trout · 2026-09-13
• Commenter: OpenAI and Anthropic's slowdown plea is really a push to regulate open source — fooobar · 2026-09-13
• AI labs' slowdown pledges split the internet: safety sincerity vs regulatory capture debate — menhguin · 2026-09-13
• Lab leaders genuinely believe in AGI risk while racing, author argues — menhguin · 2026-09-13
• Dario Amodei Gets Flak Both as a Doomer and a Regulatory Capturer, and Why That May Be Unfair — menhguin · 2026-09-13
• Dario, Musk, and Sam all back slowing AI — devs push back, citing China — gethackteam · 2026-09-13