Anthropic's Fable 5 Pulled for Foreign Nationals Amidst Government Directive and Safeguard Backlash

Anthropic’s Fable 5 and Mythos 5 models, hailed for their advanced capabilities, quickly became the epicenter of industry debate following their recent release. Initial controversy surrounded the models’ implementation of ‘invisible safeguards,’ which silently modified user prompts or steered responses when detecting queries related to frontier LLM development, without explicit user notification. Further concerns arose from a new 30-day data retention policy for Mythos-class models, which could extend to two years (and seven for classification scores) for content flagged as a usage policy violation, immediately rendering the models unusable for businesses with strict data compliance requirements like HIPAA. Following significant community backlash and criticism from researchers, Anthropic partially walked back the invisible safeguards, making them visible and falling back to Opus 4.8, admitting the initial trade-off was ‘wrong.’

The situation escalated dramatically with an abrupt US government intervention. Citing national security authorities, a directive was issued to suspend all Fable 5 and Mythos 5 access for foreign nationals, regardless of their location, including non-US citizen Anthropic employees. While Anthropic complied, they expressed strong disagreement, asserting that the single potential jailbreak technique demonstrated by the government—asking the model to identify and fix software flaws—yielded vulnerability findings widely available from other publicly deployed models, including OpenAI’s GPT 5.5. Anthropic contends that applying such a stringent standard would effectively halt new model deployment across the entire frontier AI industry. This directive creates significant operational challenges for developers and organizations, with many, including the original speaker, reporting immediate financial losses due to suspended subscriptions for their non-US-based teams.