Welcome to the AGI era of AI governance

Details

On June 13 — just after markets closed — the US executive branch ordered Anthropic to suspend access to its Claude 5 Mythos/Fable models for all foreign nationals and users abroad. The trigger: Amazon, Anthropic's largest financial and technology partner, alerted the White House to a jailbreak vulnerability in the Fable model. The situation was still unfolding when Lambert published this analysis. Analyst Nathan Lambert (Interconnects) argues the jailbreak concern was real but narrow — no frontier model is immune to jailbreaks — and the policy response was disproportionate. He also attributes partial blame to Anthropic's own rhetoric: years of comparing AI to nuclear weapons likely accelerated this style of heavy-handed intervention by 6 to 12 months. The economic contradiction is central to Lambert's critique. There is no viable domestic AI industry if foreign nationals and international developers cannot build with frontier models in the US. He also flags Amazon's escalation as puzzling — why did Anthropic's primary investor go directly to the White House rather than working through normal channels first? Lambert warns against treating this as a win for any faction. Repeated government interventions targeting the US's leading AI lab could destabilize markets. He frames this as the opening shot of the AGI governance era — one the industry entered unprepared.