Anthropic agreed to extend an existing guardrail on its Claude Fable 5 model to regain Trump administration clearance, with any user trying to unlock restricted cybersecurity or biology capabilities now notified their request is blocked and rerouted to the less-advanced Opus 4.8 model, according to Wired. The new safeguard targets a specific workaround flagged in an Amazon paper, where users bypassed a restriction by asking Fable 5 to fix code rather than identify security flaws in it, a behavior most cybersecurity experts don't find troubling but that triggered the government showdown. Commerce Secretary Howard Lutnick said Anthropic agreed to proactively detect and address model security risks, and the department cleared Fable 5 after its Center for AI Standards and Innovation judged the safeguards robust enough. Defense Secretary Pete Hegseth, however, has told advisers there's no clear path to lift his order designating Anthropic a supply-chain risk.
