The Anthropic ‘Fable’ saga proves: we have opened the AI Pandora’s box. What now? | Nathan E Sanders and Bruce Schneier
TL;DR
Anthropic released Fable on 9 June as a constrained version of Mythos. Three days later, the US government reportedly classified it as a dangerous munition. Because Anthropic could not reliably separate US users from foreign nationals, it cut off access for everyone. Schneier argues the export-control response barely addresses the risk.
Nauti's Take
The Fable saga is less proof of one mythical supermodel than proof that the control surface has moved outside the model card. Once agents can browse, write code, send mail, buy things or touch internal systems, they stop being chat products and become action layers.
Export controls may slow diffusion, but they do not replace permissions, audit logs, sandboxes and liability. The hard line for the next wave should be simple: capability without traceability is not ready for production.
Briefingshow
A ban that targets one model just buys time until the next model-harness stack reaches the same capability. For companies, the risk is no longer just the vendor name; it sits in tool permissions, agent goals and the orchestration layer around the model. Governance and security reviews need to evaluate systems, not model labels.
Sources