Anthropic unveiled Claude Fable 5 on Tuesday, a constrained version of its powerful Mythos artificial intelligence system that sparked cybersecurity concerns earlier this year. The new system carries twice the price tag of the company's previous flagship model while adding safety guardrails to block responses related to cybersecurity, biology and other sensitive areas.

The original Mythos system, released in April, prompted panic among executives and government officials who worried it could enable a new wave of cyberattacks. Anthropic initially restricted access to just 40 critical infrastructure organizations before expanding to 150 partners this month. The company's head of product management, Dianne Penn, said these security concerns drove the development of Fable's restrictions.

According to Vals AI benchmark tests, Fable outperforms all other publicly available AI systems, scoring 5 percent higher than Claude Opus 4.8. Rayan Krishnan, Vals AI's chief executive, noted Fable excels at code generation and mathematics, though other systems remain superior in healthcare and tax evaluation tasks.

Cybersecurity experts remain divided on Anthropic's cautious approach. While some praise the guardrails, others argue wider researcher access would better serve defensive purposes. Meanwhile, OpenAI has taken a different path, sharing its GPT-5.4-Cyber model with hundreds of organizations and planning broader release. The split reflects an industry wrestling with balancing innovation against security risks.