Key Points
- Federal authorities mandated Anthropic disable Fable 5 and Mythos 5, its most sophisticated AI systems, over alleged national security risks.
- The directive followed reports of a potential jailbreak vulnerability, which Anthropic characterizes as limited in scope and non-universal.
- While Anthropic followed the government’s instructions, the company has publicly challenged the decision, arguing the evidence is inadequate.
- Anthropic’s pre-IPO valuation on Hyperliquid declined 3.7% to approximately $1,627 after the announcement.
- The company committed to publishing comprehensive technical information regarding the alleged vulnerability within one day and maintains the directive results from miscommunication.
Federal authorities directed Anthropic to disable its most sophisticated AI systems on Friday, creating turbulence in pre-IPO trading and escalating tensions between the AI company and government regulators.
Details of the Federal Directive
The directive reached Anthropic at 5:21 p.m. Eastern Time on Friday. Issued through national security protocols, the mandate compelled the organization to terminate all access to Fable 5 and Mythos 5 for foreign nationals, regardless of their location.
This restriction extended even to foreign national staff members employed by Anthropic.
For complete compliance, Anthropic deactivated both systems for its entire user base. Other models in the company’s portfolio, such as Opus 4.8, remained operational.
Fable 5 and Mythos 5 had only recently become available. Both systems were developed using Mythos Preview as their foundation, a versatile model that Anthropic claimed had identified thousands of security flaws in essential software infrastructure.
Federal officials informed Anthropic they had discovered a technique capable of jailbreaking Fable 5. Such techniques enable users to circumvent the protective safeguards integrated into AI models.
Anthropics’s analysis led to disagreement. The organization concluded the technique was specific rather than universal, exploiting a limited set of previously documented, relatively insignificant vulnerabilities.
Furthermore, the company noted that alternative publicly accessible models, including OpenAI’s GPT-5.5, can identify identical vulnerabilities without requiring any workaround techniques.
“To date, the government has only given us verbal evidence of a potential narrow, non-universal jailbreak, which essentially consists of asking the model to read a specific codebase and fix any software flaws,” Anthropic said.
The organization emphasized that cybersecurity professionals routinely perform this identical function.
Market Reaction to the Shutdown
Financial markets reacted swiftly. Anthropic’s perpetual contract on Hyperliquid, which represents trader expectations for the company’s equity valuation, decreased 3.7% on Saturday, settling around $1,627.
Prior to the Fable 5 release, the contract had traded above $1,800. Current open interest stands at approximately $8.6 million, modest compared to the SpaceX perpetual contract but significant for a pre-IPO entity.
Anthropics maintains the directive stems from miscommunication and is actively working toward reinstating access expeditiously.
The organization intends to release complete technical documentation regarding the purported vulnerability within the next 24 hours.
Anthropic warned that applying this standard broadly “would essentially halt all new model deployments for all frontier model providers.”
The organization, which established its credibility through prioritizing AI safety, now finds itself openly contesting a federal directive based on its assessment that the supporting evidence fails to satisfy its own established criteria.
The outcome of this dispute—whether the directive is rescinded, modified, or expanded—will significantly influence Anthropic’s trajectory toward becoming a publicly traded company.
