Anthropic investigates report of rogue access to hack-enabling Mythos AI
9 hours ago
- #AI Security
- #Unauthorized Access
- #Cybersecurity Threat
- Anthropic is investigating reports of unauthorized access to its unreleased Mythos AI model via a third-party vendor.
- Bloomberg reported that a small group accessed Mythos using contractor credentials and cybersecurity research methods, but they are not exploiting it for attacks.
- The UK's AI Security Institute warned Mythos poses a significant cyber-threat, capable of executing complex, multi-step attacks without human intervention.
- The model's ability to autonomously find IT vulnerabilities has raised concerns among authorities about keeping such dangerous technology secure.