Anthropic said it is releasing Claude Fable 5, a safeguarded version of its high-capability Mythos AI system, to the public, despite previously warning the tool could be too powerful for general use. The company framed the rollout as controlled—citing limits and guardrails—while acknowledging the risks of making such a capable model broadly available. Mythos was previewed privately in April amid concerns from finance and government officials that it could enable hacking or system exploitation; test users reported uncovering more than 10,000 critical security flaws. Anthropic is also widening access to Mythos 5 for select cyberdefense and infrastructure groups, with fewer restrictions depending on use cases, and said both Fable and Mythos can run “unattended” for longer than prior Claude models. The move comes as U.S. agencies test Mythos even as Anthropic pursues a lawsuit involving the Pentagon’s stance on government use of its tools. With a private valuation nearing $1 trillion and expectations of going public, Anthropic’s decision underscores the industry’s race to showcase capability gains while regulators and investors weigh the attendant risks.
Related articles:
Introducing the Claude 3 model family
NIST AI Risk Management Framework
