Anthropic’s senior technical staff held in-person talks with Commerce Department officials in Washington on June 15, but the meeting concluded without agreement over the export control directive that forced global suspension of Claude Fable 5 and Mythos 5 (CNBC). Reporting published on June 16 revealed that Commerce Secretary Howard Lutnick’s letter to CEO Dario Amodei - delivered the evening of June 12 at 5:21 PM ET - gave Anthropic roughly 90 minutes to comply before the restriction took effect (BusinessToday). The directive cited the “deemed export” rule, the decades-old legal principle under which sharing technology with a foreign national in the U.S. counts as an export to that person’s home country (Fortune). Axios reported separately that Commerce acted after a second company claimed it had jailbroken Mythos, alarming administration officials, while the Lutnick letter itself provided no specific security rationale (Axios). Anthropic disputed the jailbreak claim as narrow and non-universal, noted the alleged technique applies to other publicly available models as well, and stated it continues to seek restoration of access with no committed timeline; developers calling Fable 5 or Mythos 5 API endpoints currently receive errors and are advised to fall back to claude-opus-4-8 (Anthropic).
Also emerging from the crisis is an organized industry response: more than 80 cybersecurity executives from Adobe, Zoom, Sophos, and NVIDIA signed an open letter to Secretary Lutnick and National Cyber Director Cairncross urging the administration to lift the export directive and adopt a transparent AI risk assessment process (FreeFable.org), with coverage noting that several signatories had published Cloud Security Alliance warnings about Mythos-capable cyberattacks as recently as April (EasternHerald). Separately, a breaking change now live on the Claude API requires attention regardless of the ban: setting temperature, top_p, or top_k to any non-default value on Claude Opus 4.7 or later now returns a 400 error, with Anthropic directing developers to use prompt-based techniques to guide model behavior instead (Anthropic API Docs). The same deprecation documentation adds a timing complication: claude-mythos-preview is scheduled for retirement on June 30, with claude-mythos-5 listed as the migration target - the same model currently under the export suspension - leaving teams on the preview track without a confirmed upgrade path pending resolution of the dispute (Anthropic API Docs).