The Once And Future Fable #2

ai-policyexport-controlsanthropicrule-of-lawai-governance

Abstraction: Reconstructing the 24 hours of the Fable takedown

Key points:

Timeline: Amazon (plus ~5 other companies) called administration officials Thursday night about a narrow Fable "jailbreak"; Anthropic got a 1pm Friday call to roll back Mythos/Fable with no details, then a 90-minute deadline, then export controls; Anthropic had pre-notified the government of the June 9 release with no objection.
The stated "lack of seriousness" was Anthropic asking for details rather than immediately deferring; the "Dario at a wellness retreat" claim was categorically denied by Anthropic and by Ashlee Vance (present at HQ). All demonstrated outputs are things GPT-5.5 produces without any jailbreak.
Zvi lays out three scenarios: (1) good — a panicked misunderstanding, quickly fixable; (2) bad — grievance-driven authoritarian shakedown ("pay us money or else," per Yglesias); (3) unlikely — the threat was genuinely real. Multiple sources signal it was culture-war/petty (Anthropic's cyber expert was viewed as a "radical Democrat").
Silver lining: proves the government will act at great cost when it wants to, so "the government will never do X" is dead; underscores urgency of building sane pause/takedown mechanisms before they're needed.
Cybersecurity leaders (Adobe, Zoom, Sophos CISOs via freefable.org) sign an open letter saying the move net harms US cyber defense; Dean Ball: the US now has an informal, secret, ad-hoc AI licensing regime; damages trust in the "American AI stack" and pushes allies (EU has ASML) toward alternatives.

Connections: Zvi Mowshowitz · Anthropic · Amazon · Dario Amodei · Export Controls · Rule Of Law · AI Policy