Anthropic Jun 19-21: Trump softens, access still frozen

Frame: The new signal is political de-escalation, not product restoration. Trump told Axios that he no longer viewed Anthropic and Dario Amodei as a threat, saying "Well, not now, but a week ago, maybe," and that Amodei responded "very responsibly." Axios also reported that the Commerce access controls remain part of unresolved technical talks. 1

Card 1 — what changed: The White House tone shifted after the G7 meeting, but Anthropic's own Jun 12 statement still says Fable 5 and Mythos 5 access was suspended for all customers to comply with the directive, while other Anthropic models were unaffected. 2

Card 2 — pressure building: A FreeFable open letter from cybersecurity executives and technical leaders asks the U.S. government to lift the Fable/Mythos controls and use a scientific, transparent process for future AI risk assessments. CNN's Jun 21 analysis frames the fight as evidence that U.S. AI regulation still lacks a consistent process. 34

Card 3 — the technical hinge: WSJ reports that Nicholas Carlini demonstrated Mythos finding and exploiting bugs in Ghost CMS and Linux, then became part of the Anthropic team arguing for guarded release. Anthropic's position is that the government has not shown a universal jailbreak, and that comparable bug-finding capability exists in other public models. 25

Bottom line: The event to watch is no longer just whether access comes back. It is whether Washington turns a one-off access fight into a repeatable rulebook for frontier-model cyber risk.