Key facts
- The US government ordered Anthropic to suspend access to its Fable 5 and Mythos 5 AI models for foreign nationals.
- National security concerns were cited as the reason for the suspension.
- Anthropic's pre-IPO shares fell 3.7% following the directive.
- The suspension impacts India's access to advanced AI models, raising concerns about digital equity and self-reliance.
- Experts view the move as the emergence of a new AI export-control regime, shifting from chip controls to model controls.
The US government has directed Anthropic to suspend access to its advanced Fable 5 and Mythos 5 AI models for foreign nationals, citing national security concerns. This directive led to a 3.7% drop in Anthropic's pre-IPO shares.
The suspension affects organizations that had recently gained early access to these models, including several in India across cybersecurity, telecom, finance, and banking sectors. This move has intensified calls within India for greater AI self-reliance and domestic capability building, with concerns that such restrictions could disadvantage Indian IT firms.
Experts view this development as a significant shift in how governments approach frontier AI, marking the emergence of a new AI export-control regime that extends beyond semiconductors to AI models themselves. The speed of this transition from chip controls to model controls is striking, compressing years of development into months.
Anthropic had initially kept Mythos 5 under restricted access due to its powerful cyber and reasoning capabilities, gradually expanding it through its Project Glasswing initiative before launching the more accessible Fable 5. The directive from the Trump administration, reportedly delivered with a 90-minute deadline, has halted this rollout.
Concerns around Fable 5 stem from its long-horizon autonomy and proactive self-verification capabilities, which can compress months of human penetration testing into hours, altering the risk landscape. However, some experts caution that restricting access may not be a permanent solution due to the rapid progress of open-source models.