US export order suspends Anthropic models; crypto AI tokens jump

The US government issued an emergency export control order that required Anthropic to suspend global access to its frontier models, Fable 5 and Mythos 5. The directive covered all foreign nationals, including the company’s international staff, and Anthropic disabled the models for its entire global user base.

US officials cited a national security vulnerability that could let users bypass the models’ safety guardrails. Anthropic’s engineers characterized the exploit as narrow, saying it allowed the models to analyze specific codebases and identify minor, previously known software flaws.

Anthropic reported that its teams performed thousands of hours of red-teaming before Fable 5’s commercial launch with participation from US and UK safety bodies and independent contractors. The company wrote that perfect immunity to jailbreaks is not currently possible and added that it designs safeguards to make any jailbreaks narrow or expensive to produce while pairing them with monitoring. Anthropic argued that applying a zero-exploit standard across the private sector would halt frontier model development and criticized the enforcement action for lacking transparency and clear technical grounding.

The enforcement action is the first time Washington has effectively restricted a widely deployed commercial frontier AI model, according to industry observers. Security practitioners have used models like Mythos to locate code-level vulnerabilities and to find and fix flaws before attackers exploit them.

Markets reacted within hours. Tokens tied to decentralized compute and open AI infrastructure rose: Bittensor’s TAO gained about 13.4%, Venice Token increased roughly 18%, and Internet Computer advanced about 9.8%. Trading firms and protocol developers said the order created an incentive to accelerate decentralized alternatives that avoid single points of control.

A Bittensor wallet and infrastructure provider wrote, “We are not building decentralized AI because it sounds better. We are building it because the off-switch cannot belong to one hand.” A Chainlink executive wrote that when intelligence runs through centralized chokepoints, access can change overnight and argued for decentralized models and verifiable infrastructure to connect and secure them.

A venture investor described the clash between Anthropic and regulators as creating momentum for decentralized AI infrastructure in real time. Another prominent investor warned that expanding compliance requirements could create heavy institutional burdens for startups while also acknowledging the need for federal oversight to reduce risks to critical infrastructure.

Analysts noted that models able to find software flaws are used by defenders and researchers. Asset managers earlier this year identified AI-linked tokens as relatively resilient during market weakness, and today’s price moves reflected renewed investor interest in decentralized architectures that operate outside a single company or jurisdiction.