Anthropic Halts AI Models Following US Government Directive
On June 12, Anthropic ceased public access to Claude Fable 5 and Mythos 5 after the US Commerce Department mandated restrictions on access from foreign nationals.
However, a select cadre of approximately 200 organizations engaged in Project Glasswing retain access to Mythos Preview, an experimental adaptation of the model aimed at bolstering cybersecurity research efforts.
The emergency export-control order, executed by Commerce Secretary Howard Lutnick, marks an unprecedented federal intervention in the cessation of an operational commercial AI system.
In light of this directive, Anthropic opted for a global deactivation of both models, as their platform lacks the capability to authenticate user nationality in real-time.
Entities such as Cisco Systems, Dragos, AWS, and JPMorgan Chase, partaking in Project Glasswing, have maintained access to Mythos Preview, utilized to identify software vulnerabilities.
Jon Lavender, CTO of Dragos, confirmed ongoing access, a sentiment echoed by Cisco. This scenario conjures an intriguing paradox: while the more advanced iteration of the model—capable of autonomously detecting flaws in operating systems and browsers—remains operational for private networks, the publicly facing counterpart has been withdrawn.
The European cybersecurity agency, ENISA, which had been invited to join Project Glasswing prior to the government’s intervention, was informed last week that access would no longer be granted.
This development indicates that Anthropic is reevaluating participation based on adherence to US export regulations, though the specific criteria have not been publicly disclosed.
The origins of this conflict date back to July 2025 when the Pentagon awarded $200 million in contracts for AI technologies to Anthropic, OpenAI, Google, and xAI.
Tensions escalated following the deployment of Claude during a special operations raid in Venezuela in January 2026.
Anthropic sought assurances that Claude would not be employed in fully autonomous weapon systems or for mass surveillance purposes. The Pentagon rebuffed these demands, prompting Anthropic to initiate legal proceedings.
In an effort to sidestep the military stalemate, Anthropic commercially launched Fable 5 and Mythos 5 on June 9. Merely three days later, at 1:00 p.m. ET, the White House issued a warning, stating that the models constituted a national security risk, demanding their disablement within a tight 90-minute window.
The catalyst for this directive: Amazon researchers reportedly circumvented safety filters on Fable 5, with the NSA affirming that the protective measures could be dismantled. Amazon’s CEO Andy Jassy contacted Treasury Secretary Scott Bessent directly to convey these findings.
High-level discussions commenced on June 15 in Washington, with Anthropic representatives engaging with officials from the Commerce Department.
Indications from the Commerce Department suggest a potential reopening of Fable 5 for commercial operations, contingent on the resolution of identified jailbreak vulnerabilities to the administration’s satisfaction.
The White House has indicated that any forthcoming private AI model surpassing current performance metrics must endure government scrutiny prior to public deployment.
For an industry characterized by rapid, global rollouts, this sets a significant precedent: a federal agency now possesses the authority to unilaterally withdraw a trillion-dollar product over a mere weekend.
Source link: Technobezz.com.
