Anthropic has launched Claude Fable 5, a more secure version of its Claude Mythos 5 AI model, alongside a new set of safeguards aimed at limiting misuse. The company says the latest model is far more capable than its earlier systems, making stronger safety measures necessary. As part of the rollout, Anthropic has placed strict restrictions on discussions related to cybersecurity, biology, and chemistry, areas that could be linked to harmful activities. Whereas Claude Fable 5 could previously provide a direct response to these sorts of sensitive questions, it is now being replaced by a less advanced model that was built earlier. This decision follows increasing awareness in the AI industry of both technology and safety.
In a blog post, Anthropic said the most significant changes in Claude Fable 5 relate to how it handles sensitive subjects. The company explained that the chatbot may decline to answer certain queries involving cybersecurity, biology, or chemistry. In some cases, those requests may instead be handled by Claude Opus 4.8.
Also read: Salesforce lays off more employees across teams, AI to takeover: Report
Anthropic’s decision is driven by the growing capabilities of its latest AI systems. Claude Fable 5 is built on the same foundation as Claude Mythos 5, the company’s most advanced AI model, which is currently available only to a limited group of trusted users.
Mythos 5 has shown notable improvements in cybersecurity-related tasks. Researchers recently used the model to analyse multiple lines of code and identify software vulnerabilities in macOS. These capabilities demonstrate the potential benefits of advanced AI systems, but they also raise concerns about how such tools could be misused.
Anthropic says it is taking a cautious approach because highly capable AI models could enable users to perform activities that would otherwise require specialised expertise. Therefore, the company has decided to add more safeguard measures on sensitive topics as they continue to evaluate the impact of the technology.
Anthropic acknowledges that information useful to researchers and cybersecurity professionals could also be dangerous if it falls into the wrong hands. For that reason, access to Mythos 5 remains restricted through Project Glasswing, a programme designed for approved cybersecurity experts.
The company plans to expand trusted access in the future, including to certain life sciences organisations. For now, however, Anthropic says it is prioritising safety, reflecting the wider debate across the AI industry about how increasingly powerful AI systems should be deployed and shared with the public.