Anthropic has introduced AI models Claude Fable 5.2 and Mythos 5.2 for vulnerability detection. In the CVE Rediscovery test, Mythos 5.2 identified 89.6% of known vulnerabilities, compared to 78.3% for the previous version.

Access: Mythos for Verified Users, Fable for All

Fable 5.2 retains restrictions on exploit creation and penetration testing and is available to all users. Mythos 5.2 has relaxed these restrictions, so access is granted only to verified organizations and specialists.

Unique Findings and Firefox Performance

At one company, every sixth vulnerability found by Mythos 5.2 was missed by all other advanced AI models in the comparison. When searching for vulnerabilities in Firefox, Mythos 5.2 showed roughly double the improvement over the previous version.

Risks for Smart Contracts and DeFi

Further development of such models could significantly accelerate the discovery of vulnerabilities in smart contracts and DeFi protocols, including by malicious actors.