Anthropic's Fable 5 Blocks Dangerous Topics to Prevent Misuse

Anthropic has publicly released Claude Fable 5, a powerful new AI model. The company has stated that this model surpasses its earlier Opus models in overall capabilities. However, Anthropic has also added strict safeguards. Fable 5 will not answer queries about cybersecurity, biology, or chemistry. The company has expressed concerns that malicious actors could use the model to cause serious harm. Fable 5 operates on the same underlying technology as Mythos 5, a more advanced model that is only available to a small group of trusted cyberdefenders. The public version of Fable 5 uses a classifier system to detect dangerous prompt subjects. When the system detects a banned topic, it redirects the query to the older Claude Opus 4.8 model and warns the user. Anthropic has admitted that these safeguards are 'stricter than ideal.' The system occasionally blocks harmless requests in less than five percent of sessions. The company has said these false positives are worth it to avoid serious risks. External red-teaming teams have tried to break the safety rules for over 1,000 hours. They have not found any universal jailbreak. The company has also reported that Mythos 5 scored 78 percent on the ExploitBench test, a large jump from Opus 4.8's 40 percent. Anthropic has started to expand access to trusted cybersecurity professionals through Project Glasswing. It will also include life sciences organizations. API users can access Fable 5 for $10 per million input tokens and $50 per million output tokens. These prices are 67 to 100 percent higher than those for OpenAI's GPT-5.5.