Anthropic Disables Fable 5 and Mythos 5 After Export Controls
Anthropic disabled access to its Fable 5 product and the underlying Mythos 5 model for all users worldwide on Friday, June 12, 2026 after the United States Commerce Department imposed national security export controls barring the company from distributing the models to foreign nationals, Fortune reported in its Sunday update at 5:10 p.m. Eastern. The Commerce Department action followed a Thursday warning from Amazon researchers who had used prompts to extract information about cyberattacks from the Mythos 5 model that was supposed to be restricted, according to the Fortune account.
Senior technical staff from Anthropic were dispatched to Washington over the weekend for in-person meetings with administration officials, with the high-stakes resolution sessions having begun virtually on Saturday and continuing into formal meetings on the ground in Washington through the early part of the week. White House AI Adviser David Sacks said publicly that the administration had asked Anthropic Chief Executive Officer Dario Amodei to, in his characterization,
fix the issue or withdraw the model
with the Friday afternoon escalation sequence taking the second of those two options off the table for the company.
The structural shape of the Friday escalation is the operational element the policy press has been parsing across the weekend. Amazon researchers raised the cyberattack-information concern to the administration on Thursday, several calls between Amodei and senior administration officials followed across Thursday evening and Friday morning, and by Friday evening the Commerce Department had stepped in with national security export controls that effectively required the company to pull the model from public access. Anthropic's response, the decision to disable both Fable 5 and the underlying Mythos 5 model globally rather than attempt a geofenced or access-restricted rollback, was the calendar-engineering move the company's leadership made through the Friday late-evening sequence.
The Amazon safety angle is the policy-narrative point the Sunday Fortune piece treats as the operational trigger for the administration response. The cyberattack-information extraction that Amazon researchers reportedly produced is the kind of guardrail failure the model safety community has been tracking through the spring of 2026 as the operational concern most likely to draw an export-control response. The mechanism Amazon researchers used, prompts engineered to bypass the model's restricted-information policies, is the failure mode the Anthropic safety team has been working against across the company's model-card revisions and post-training adjustments. Whether the failure resulted from a routine policy gap that has now been closed in the model's post-Friday version or from a more structural issue requiring a longer model-retraining cycle was the question Anthropic's Washington meetings opened.
The Treasury and Commerce Department roles are the inter-agency dimension the resolution conversation will need to resolve before either model returns to public access. Treasury Secretary Scott Bessent had, according to the Fortune reporting, suggested to Anthropic leadership that the company was making the wrong decision in disabling the models globally rather than negotiating a foreign-national access restriction within the existing model deployment, while the Commerce Department's national security export controls were the operative legal mechanism that produced the Friday evening shutdown. The intra-administration positioning between the two departments, with Commerce taking the harder enforcement posture and Treasury suggesting a softer middle ground, is the politics the Anthropic leadership team will be navigating in the Washington meetings.
The economic and product implications run on two tracks for Anthropic's customer base and for the broader commercial AI market. The first track is the immediate revenue impact, with the Fable 5 product representing the company's flagship consumer-and-developer offering and Mythos 5 the underlying model used across the company's enterprise and API customer segments. The disable-globally decision pulls both offerings out of active commercial circulation while the Washington conversation continues, and the company's earlier model versions remain available to fill the access gap. The second track is the broader commercial AI market positioning, with the export-control action setting an early precedent for how the Trump administration will use existing trade-control mechanisms to manage advanced model access. OpenAI, Google, and the smaller frontier-model labs will be watching the Anthropic resolution as the template for what to expect if their own next-generation models trigger similar safety concerns.
What sits ahead in the immediate week is the resolution of the Washington meetings and whether the Commerce Department's export-control determination is modified to permit a restricted or geofenced reactivation of either model. The Anthropic team's stated position, that the White House did not raise Chinese access to Mythos in its conversations with the company per the company spokesperson quote in the Fortune piece, suggests the safety-versus-export-control framing of the administration response is the point the company will be working to shift in the early-week negotiations. Whether the models return in their current configuration, return in a modified version, or remain offline through a longer policy-and-engineering review cycle will be settled in the next several days, with the broader implications for the United States commercial AI sector folded into whichever resolution the parties reach.
