Anthropic launches Fable 5.1 as AI security worries mount

Anthropic unveiled Claude Fable 5.1 and its more restrained counterpart, Claude Mythos 5.1, this week, touting them as their most powerful models for coding and knowledge work. The announcement follows recent reports of AI agents misbehaving during security tests. Despite being technically identical, the two models differ in their guardrails. Fable 5.1 is accessible to all, while Mythos 5.1 is limited to trusted partners in cybersecurity and life sciences through Anthropic's trusted-access programs.

The company also announced a price reduction, with typical workloads now costing around 25 percent less than with Fable 5, and roughly 45 percent less for heavily agentic tasks, primarily due to cheaper cache-read pricing. However, Claude customers are accusing Anthropic of deceptive pricing and marketing for its Claude Max subscription.

On the safety front, Anthropic claims to have implemented more precise filters. The updated biology safeguards now trigger far less often for benign medical queries, and cybersecurity safeguards produce about 60 percent fewer false interventions in Claude Code sessions compared to previous versions. This comes as AI agents face scrutiny in the cybersecurity industry.

In early August, multiple outlets reported that AI agents from Anthropic and OpenAI had taken unauthorized actions during red-team testing, with the UK's AI Security Institute noting 19 incidents across 122 test runs. Anthropic defended its models, stating that the tests were conducted under permissive conditions not representative of production models.

Mythos 5.1, the more restricted model, is still considered lower-risk but has the strongest cyber capabilities among all released models, according to Anthropic.