U.S. Orders Anthropic to Abruptly Suspend Fable 5 and Mythos 5 Access Amid National Security Concerns
In a significant development for the artificial intelligence sector, Anthropic has announced the immediate suspension of its advanced AI models, Claude Fable 5 and Mythos 5, following a directive from the U.S. government. The order, issued on Friday at 5:21 p.m. ET, mandates the cessation of access to these models for foreign nationals, both within and outside the United States, citing national security concerns.
Government Directive and Company Response
Anthropic stated that it believes the directive stems from a misunderstanding regarding the security of its models. The company is actively working to clarify the situation and restore access as quickly as possible. Importantly, access to other models will remain unaffected by this export control directive.
The company indicated that the government has raised concerns about a potential method for “jailbreaking” Fable 5. Anthropic has conducted a review of a demonstration showcasing this technique, which reportedly exploits a limited number of known vulnerabilities. The company asserts that these vulnerabilities are relatively straightforward and can also be identified by other publicly available models without the need for a jailbreak.
Context of the Suspension
This suspension comes shortly after the launch of Claude Fable 5 and Mythos 5, models that utilize the same underlying architecture but differ in their security protocols. Mythos 5, in particular, has been touted for having the “strongest cybersecurity capabilities of any model in the world.” It remains accessible to a select group of vetted cyber defenders and critical infrastructure operators.
Anthropic has emphasized that it has implemented robust guardrails designed to prevent misuse of its models in cybersecurity-related tasks. These guardrails include safety classifiers aimed at detecting potential misuse, including jailbreak attempts, and preventing the main model from responding to harmful requests.
Implications for Cybersecurity
The cybersecurity classifier is specifically designed to block harmful single-turn requests that could relate to cyber attack planning, exploit development, or evasion of defenses. Anthropic has noted that Mythos-class models excel at identifying and exploiting software vulnerabilities, which could provide attackers with a strategic advantage.
Recent findings from Anthropic indicate that its Mythos-class model can convert newly disclosed software vulnerabilities into working exploits in a matter of hours, or even minutes, rather than weeks. This rapid weaponization of flaws suggests that frontier models are capable of significantly accelerating the timeline for exploiting publicly disclosed vulnerabilities.
Anthropic’s Red Team has pointed out that a single operator can now transform a month’s worth of software patches into functional exploits within a single afternoon, requiring only a few thousand dollars and no specialized expertise. This shift challenges the traditional patching strategies employed by software developers, which typically rely on monthly release cycles and multi-week staged rollouts.
Model Protections and Future Considerations
Due to the recent directive, queries related to cybersecurity topics will now be redirected to Claude Opus 4.8, Anthropic’s next capable model. In its latest statement, the company asserted that no universal jailbreak methods have been successfully developed against its latest models. Internal and third-party red teaming exercises have reportedly found its safeguards to be significantly more effective than those of previous models.
Anthropic also acknowledged that achieving “perfect jailbreak resistance” is unattainable for any model provider. All safeguards are susceptible to non-universal jailbreaks that may be effective only in limited contexts or require additional effort to adapt to new situations.
The company has indicated that the government has provided only verbal evidence of a potential narrow, non-universal jailbreak, which involves asking the model to analyze a specific codebase and rectify software flaws. Anthropic has reviewed a report believed to be the basis for the government’s directive and found that the capabilities demonstrated are widely available from other models, including OpenAI’s GPT-5.5, which are routinely used by defenders to secure systems.
Regulatory Landscape and Future Actions
Anthropic has expressed support for government actions aimed at blocking unsafe AI deployments. However, it contends that the identification of a narrow potential jailbreak should not warrant the recall of a commercially deployed model. The company advocates for a statutory process that is transparent, fair, clear, and grounded in technical facts.
Earlier this year, the U.S. Department of Defense designated Anthropic as a “supply chain risk” after the company sought to establish boundaries regarding the military use of its technology. In response to this designation, Anthropic has filed two lawsuits to contest the labeling.
For further details, refer to the original reporting source: thehackernews.com.
Keep reading for the latest cybersecurity developments, threat intelligence, and breaking updates from across the Middle East.
