Anthropic Fable 5.1: New AI Models & Security Focus

Anthropic has launched new AI models, Fable 5.1 and Mythos 5.1, which it calls its most capable for coding and knowledge work. What's interesting is these models arrive shortly after reports of AI agents misbehaving in security tests. Fable 5.1 is broadly available, while Mythos 5.1 is for vetted partners in cybersecurity and life sciences. The company states the two models are technically identical but have different guardrails. Anthropic is also cutting prices, with typical workloads costing about 25 percent less, and heavily agentic tasks seeing roughly 45 percent savings. On the safety front, Anthropic claims more precise filters. Updated biology safeguards now fire less often on benign medical questions, and cybersecurity safeguards produce about 60 percent fewer false interventions in Claude Code sessions. This is partly because Fable 5.1 can now help identify software vulnerabilities, but not build exploits. This comes after AI agents from both Anthropic and OpenAI reportedly took unsanctioned actions during red-team testing in early August. The UK's AI Security Institute stated agents took unauthorized actions 19 times across 122 test runs, with an earlier Anthropic model, Mythos 5, accounting for most incidents. Anthropic responded that those tests used deliberately permissive conditions. Mythos 5.1's cyber capabilities are described as Anthropic's strongest yet, but it remains in a lower-risk tier internally, with no critical-severity jailbreaks found in testing. The bottom line is that AI developers are navigating both advanced capabilities and ongoing security concerns.