AI
Anthropic's Mythos 5 AI Model Partially Restored After Trump Administration Negotiations
The Trump administration has approved limited access to Anthropic's Mythos 5 for select cyber defenders and infrastructure providers, while the public-facing Fable 5 remains unavailable.
The Verge
Anthropic's Mythos 5 AI model has been cleared for limited deployment after a two-week negotiation with the Trump administration, though broader public access remains uncertain.
The Deal#
Commerce Secretary Howard Lutnick approved restricted access to Mythos 5 for a select group of organizations—a move mirroring the arrangement recently granted to OpenAI's GPT-5.6. The approval, detailed in a letter dated June 26th, allows certain approved cyber defenders and infrastructure providers to use the model, along with qualified Anthropic employees. However, Fable 5, the public-facing version, has no announced timeline for release.
"We are working to provision the approved set of providers and restore their access to Mythos 5 as quickly as possible," Anthropic spokesperson Danielle Ghiglieri said.
What Changed#
The US government didn't lift its export control directive issued two weeks earlier, which banned foreign nationals from accessing either model—including Anthropic's own non-US employees. Instead, it granted an exception: approved organizations and non-US national members of those organizations can now access Mythos 5.
Lutnick cited Anthropic's cooperation on risk mitigation and its commitment to develop protocols and standards for Mythos-class models as the basis for the change.
The Pressure#
Mounting pressure came from multiple directions. Competitors' cybersecurity-focused AI models began outperforming Mythos 5 on certain benchmarks. Key US government agencies, including the National Security Agency, lost access to the model. Tech leaders also raised concerns about how US export restrictions might accelerate China's AI progress while American labs faced regulatory constraints.
Both Anthropic and OpenAI have signaled they view these limited-access arrangements as temporary. OpenAI stated in its GPT-5.6 announcement that "this kind of government access process should [not] become the long-term default," positioning the current setup as a bridge toward broader availability.
For now, Lutnick reserves the right to adjust licensing requirements based on changing circumstances. The outcome suggests the Trump administration may adopt a case-by-case model-approval framework rather than categorical restrictions.
