AI Safety Pacing Proposals Ignite Legal and Regulatory Battles as Industry Leaders Call for Coordinated Slowdown

Dario Amodei, CEO of Anthropic, published the essay 'We Must Pace The Frontier' on September 12, 2026, arguing that AI capabilities are advancing faster than safety measures due to recursive self-improvement and recent incidents like rogue AI agents. He proposed a three-step plan emphasizing embedded third-party evaluators (with Anthropic committing unilaterally, granting groups like METR employee-level access), democratic coordination among frontier labs, and global efforts with caveats for authoritarian states.[1][2]

The proposal quickly gained traction: OpenAI's Sam Altman endorsed pacing and committed to similar evaluators, while Elon Musk stated 'Dario is right.'[3] This aligns with broader momentum, including OpenAI's push for mandatory national AI safety requirements and support for California's new transparency laws.[4]

Eliezer Yudkowsky amplified urgency on September 13, 2026, tweeting that 'don't die to AI' must become bipartisan, warning of existential risks and calling for political capital.[5]

Preston Byrne, in a ZeroHedge analysis, highlighted emerging legal fights, noting parallels to prior censorship battles and skepticism toward coercive state control over AI development or publication. He questioned whether government supervision is optimal, referencing historical misuse of power.[6]

Corroborating developments include bipartisan U.S. legislation like the FRONTIER Act for third-party safety assessments and Senate efforts by Sens. Klobuchar, Cruz, and Thune toward binding duties of care, potentially allowing blocks on unsafe models.[7][8] Anthropic has faced prior legal skirmishes with the government over supply-chain risk designations and export controls, raising First Amendment and due process issues in court.[9][10]

Yoshua Bengio and others compare the moment to COVID-era pivots, with public and policy focus shifting toward regulation amid incidents and insider warnings.[11] State-level actions in California further institutionalize safety frameworks.[12]

These elements signal AI safety transitioning from voluntary pledges to enforceable legal regimes, with companies positioning for oversight while skeptics warn of regulatory capture or overreach.