What happened
On June 26, 2026, Commerce Secretary Howard Lutnick sent a letter to Anthropic stating that 'appropriate safeguards are in place to permit certain trusted partners to access the Claude Mythos 5 Model,' removing the export-license requirement for more than 100 named US companies, federal agencies, and their foreign-national employees. The companion model Fable 5 remains fully suspended under the June 12 export-control directive. The letter establishes a formal trusted-partner annex (not publicly disclosed) and explicitly covers deemed-export and in-country-transfer scenarios.
Why it matters
This is the first time the US government has used Export Administration Regulations to gate access to a commercial AI model and then selectively re-authorised it via a named-entity list — creating a de-facto licensing regime for frontier AI. It signals that any AI model deemed to have national-security-relevant cybersecurity capability can be suspended and conditionally restored by a single Commerce Secretary letter, with no public criteria, rulemaking, or appeals mechanism. Enterprises relying on frontier AI APIs now face a new category of access-interruption risk: federal order, not just vendor outage or pricing.
Action needed
Organisations using or planning to use Anthropic models should (1) verify whether they appear on Annex A; (2) assess supply-chain dependency on models that could be suspended without notice; (3) review contracts for force-majeure or government-order clauses. Fable 5 access remains blocked — no restoration timeline published.