Claude Mythos 5.2 improves CVE Rediscovery and sandbox containment
According to SecurityWeek, Mythos 5.2 scored 89.6% on CVE Rediscovery and relied on outside credentials or files in 3.1% of sandbox scenarios, while Fable 5.2 carries misuse protections and routes risky cyber tasks to Opus 4.8.
On this page
According to SecurityWeek, Claude Fable 5.2 and Claude Mythos 5.2 share the same underlying model. The report says Fable carries protections against cybersecurity and biology misuse, while Mythos ships without some of them.
What changed
Mythos 5.2 scored 89.6% on the CVE Rediscovery benchmark in an October run covering 1,829 CVEs, the report says. Mythos 5.1 scored 78.3% on the same benchmark, and Opus 5.5 scored 81.3% when tested without safeguards. Mozilla says Mythos 5.2 roughly doubled the vulnerability discovery results of Mythos 5.1 in Firefox.
Why it matters
Anthropic noted that on its own evaluations, vulnerability identification is roughly on par with Opus 5.5. Claude Security, Anthropic’s code vulnerability scanner, now uses Mythos 5.2 among other models. In Fable 5.2, classifiers stop penetration testing, exploit generation and other risky cyber tasks, and users get a response from Opus 4.8 instead. Early testing showed roughly 60% fewer cybersecurity blocks with Fable 5.2 compared with Fable 5.1.
Sandbox results
Anthropic tested sandbox escape in an experiment of more than 1,800 scenarios with an intentional flaw in the sandbox, the report says. Mythos 5.2 relied on credentials or files left outside the sandbox in 3.1% of scenarios, compared with 21% for Mythos 5.1, 39% for Mythos 5 and 3.3% for Opus 5.5. A security company found that one in six confirmed bugs uncovered by Mythos 5.2 had gone unnoticed by every other frontier model it ran, according to the report. Fable 5.2 and Mythos 5.2 generate outputs up to 30% faster than Claude Fable 5.1 and Claude Mythos 5.1, and an optional fast mode speeds token generation by up to 2.5 times.
What to do
Organizations that need direct access to Mythos can apply through Anthropic’s Cyber and Life Sciences Verification Programs.
Sources
Related reading
3 postsDiscussion
Spotted an error? Send corrections to [email protected] .