Fable 5 can discuss biology again: Anthropic significantly relaxes overzealous safety filters

Anthropic has reduced the false alarm rate of its biology safety filters for its Fable 5 model by approximately 85 percent. Previously, nearly all biology-related queries were blocked and rerouted to the less capable Opus 5 model, causing protest from scientists. Now, users can handle significantly more biology tasks with Fable 5, such as interpreting laboratory results, understanding symptoms, or answering medical questions. For so-called dual-use topics like virology, toxicology, or molecular design, the block remains in place, as Anthropic claims Fable 5 could provide malicious actors with capabilities they cannot find elsewhere. The company is working on access programs to give researchers access to restricted functions in the future. The concern behind these safety measures is that biological risks could be more dangerous than cyberattacks because they are harder to defend against: once a virus is released, it cannot be easily recaptured, and a cure may take a long time. A threat analysis by US intelligence agencies, an evaluation of ChatGPT chats with bioweapon requests, joint warnings from researchers, and a Stanford project that designed fully synthetic viruses with AI confirm that the threat may be real and not just typical Anthropic fear.