Anthropic has introduced a new AI system that is exceptionally skilled at finding software bugs and security vulnerabilities. The company admits the technology is so capable that, if misused, it could pose significant cybersecurity risks.
To address these concerns, Anthropic split the technology into two versions. Claude Fable 5 is the public model, offering advanced coding, analysis and vision capabilities while including safeguards that prevent it from actively searching for dangerous exploits. Claude Mythos 5, which is equally powerful but has fewer restrictions, will only be available to trusted cybersecurity researchers.
Anthropic has also developed specialized monitoring systems designed to detect and block harmful activities, such as discovering zero-day vulnerabilities, creating dangerous chemical or biological compounds or attempting to replicate the model without its built-in safety measures. The company hopes this approach will allow the AI’s advanced capabilities to improve security while minimizing the risk of misuse.
Source: Android Authority
