The Commerce Department has lifted the export controls on Claude Fable 5 and Mythos 5, ending the eighteen-day suspension that began June 12. Anthropic announced the decision today; CNBC reports the department's letter cites a set of coordination commitments Anthropic made to the government. Fable 5 itself returns to service tomorrow, July 1 — details in our lead story.
One detail worth noting: the negotiations were led by co-founder Tom Brown, not CEO Dario Amodei. Brown is the technical co-founder most associated with Anthropic's compute and scaling work, and putting him across the table reads as a choice to make the conversation an engineering one — what safeguards exist, how they are tested, who gets to verify them — rather than a policy standoff.
The four commitments
Commerce's letter rests on four commitments, and they are worth reading as the template they will likely become:
1. Pre-release government access. Designated government partners get access to frontier models before release, for capability evaluations and guardrail testing.
2. Rapid jailbreak response. Anthropic will investigate, triage, and notify the government of significant jailbreaks or misuse, and share new safeguards for independent testing — the process that was ad hoc when the Amazon researchers' report landed in June is now formalized.
3. Dedicated research resources. Anthropic teams working on government priorities, backed by significant compute and red-teaming expertise.
4. Industry standards. Shared voluntary security and evaluation standards for frontier model providers — not just Anthropic.
None of these is a one-time concession; all four are standing obligations. In effect, Anthropic bought back global distribution by wiring the government into its release process. Whether the other frontier labs adopt the same terms voluntarily, or wait to be compelled, is now one of the more consequential open questions in the industry.
The same-day Sonnet 5 launch
Anthropic did not spend the day only on regulatory news. It also launched Claude Sonnet 5 (claude-sonnet-5), the new default model for Free and Pro users, at introductory pricing of $2 per million input tokens and $10 per million output through August 31, 2026 — then $3/$15. On Anthropic's agentic coding benchmark it scores 63.2% against Opus 4.8's 69.2% and Sonnet 4.6's 58.1%; TechCrunch frames it as a cheaper way to run agents, and VentureBeat notes the steep discount against the top of the line.
The timing is not subtle: on the day the government cleared its most capable model to return, Anthropic shipped its cheapest competent one. The lineup that greets returning users this week spans $2/$10 to $10/$50 per million tokens.
What to do about it
If you shelved Fable 5 work on June 12, dust it off tonight — global access returns July 1 and the model id is unchanged. If you maintain migration docs or default-model settings, note that Sonnet 4.6 advice now points to Sonnet 5, and the $2/$10 intro window runs through August 31. And if you write internal AI policy, read the four commitments closely: they are the likely shape of frontier-model regulation, arriving as a letter rather than a law.