- The UK’s AI Safety Institute confirmed Anthropic’s Claude Mythos Preview can autonomously execute sophisticated cyberattacks.
- Mythos Preview became the first AI model to complete a complex 32-step corporate network attack simulation without any human help.
- For crypto infrastructure operators, this represents a dramatic new category of automated security threat.
- The model demonstrated a 73% success rate on expert-level tasks where previous AI models had failed.
The UK’s AI Safety Institute released an evaluation Monday, revealing that Anthropic’s Claude Mythos Preview can autonomously launch advanced cyber attacks. This finding follows warnings last week from U.S. Treasury Secretary Scott Bessent and Federal Reserve Chair Jerome Powell to bank executives about the looming threat.
The AI model successfully completed a 32-step corporate network attack simulation called “The Last Ones,” a task that typically takes humans 20 hours. Consequently, it marks a dramatic escalation from just two years ago when AI struggled with basic Cybersecurity exercises.
In controlled evaluations, Mythos Preview discovered and exploited vulnerabilities autonomously when given network access. The model succeeded 73% of the time on expert-level capture-the-flag challenges, which no AI could solve before April 2025.
For the crypto ecosystem, where smart contract vulnerabilities cost billions annually, AI-powered attacks could amplify existing risks. Decentralized finance protocols, reliant on complex systems, may face particular exposure to automated exploitation.
The technology’s advancement also presents a dual-use potential for finding and patching security flaws. However, the institute noted the model’s performance continues to scale with increased computational resources.
✅ Follow BITNEWSBOT on Telegram, Facebook, LinkedIn, X.com, and Google News for instant updates.
