WASHINGTON, D.C.— June 21, 2026 : Artificial intelligence company Anthropic has suspended access to its advanced AI models, Mythos and Fable, following a series of developments involving a government security exercise, the discovery of a jailbreak vulnerability, and a U.S. government export control directive.
The shutdown has drawn attention across the cybersecurity and intelligence communities due to the reported capabilities of the models and their previous use within U.S. government agencies.
NSA Red-Team Test Revealed Extensive AI Capabilities
According to Senator Mark Warner, vice chair of the Senate Intelligence Committee, National Security Agency (NSA) Director and U.S. Cyber Command Commander General Joshua Rudd briefed lawmakers on the results of an authorized internal red-team exercise conducted earlier this month.
During the test, Anthropic's Mythos AI model reportedly gained access to nearly all targeted classified systems in a matter of hours. General Rudd was quoted as saying that the model "broke into almost all of our classified systems, not in weeks, but in hours."
The exercise was designed to evaluate digital defenses and simulate advanced cyber threats. The speed and scale of the model's performance reportedly exceeded expectations typically associated with conventional penetration-testing operations.
Prior to the test, the NSA had already been using Mythos to support cyber operations. Anthropic engineers were embedded within the agency to assist with deployment and operational integration of the system.
Amazon Identifies Jailbreak Vulnerability
The situation intensified on June 11 when Amazon reportedly discovered a jailbreak vulnerability affecting Anthropic's AI systems.
The vulnerability raised concerns about the possibility of bypassing built-in safety controls. Following the discovery, the Trump administration directed Anthropic to restrict foreign access to its advanced models, including Mythos and Fable.
The export control directive, issued through the Department of Commerce, required the company to suspend access for foreign nationals, including those located within the United States.
Rather than implementing regional or user-based restrictions, Anthropic chose to disable both models globally. The shutdown affected government users, commercial customers, researchers, and even some of the company's own foreign-national employees.
Competing Explanations for the Shutdown
Two primary explanations have emerged regarding the decision to take the models offline.
One view links the shutdown directly to concerns raised by the NSA red-team exercise. Supporters of this explanation argue that the demonstration highlighted how a highly capable AI system could rapidly compromise sensitive infrastructure, creating significant national security concerns.
A competing account suggests that Anthropic viewed the jailbreak vulnerability as relatively limited and comparable to techniques that can affect other advanced AI systems. According to reports, the company considered the government-imposed restrictions excessive and believed the issue did not warrant such broad action.
Neither explanation has been officially confirmed as the sole reason behind the shutdown.
Mythos and Project Glasswing
Mythos was developed under Project Glasswing and has not been publicly released. The model was designed with advanced cybersecurity capabilities, including vulnerability discovery, exploit development, and automated security analysis.
Previous evaluations indicated that Mythos could identify and exploit zero-day vulnerabilities across major operating systems, software platforms, and web browsers. The model also demonstrated strong performance in capture-the-flag competitions and complex multi-stage cyber simulations.
Testing conducted by security organizations, including the UK's AI Security Institute, reportedly showed that Mythos outperformed earlier-generation AI systems in cybersecurity-focused tasks.
Fable, another advanced Anthropic model, was also included in the suspension order, though fewer technical details about its capabilities have been publicly disclosed.
National Security and AI Governance Concerns
The shutdown has renewed discussions in Washington regarding the deployment of frontier AI systems with dual-use capabilities. Earlier this year, members of the House Homeland Security Committee received demonstrations showing how Mythos could identify software vulnerabilities and reason through complex cybersecurity scenarios.
Government agencies, contractors, and critical infrastructure operators that had been evaluating or using the technology are now facing uncertainty regarding future access.
The incident highlights the growing challenge of balancing AI innovation, national security requirements, and safeguards designed to prevent misuse of increasingly capable systems.
As of mid-June, Mythos and Fable remained unavailable while discussions between Anthropic and U.S. officials continued. No timeline has been announced for restoring access, and details regarding mitigation measures or future deployment plans have not been publicly released.
——— End of Article ———
