Quoting Anthropic Frontier Red Team

Blog · Watch/Read · (gathered 2026-09-30)

• Simon Willison highlights a critical finding from Anthropic’s Frontier Red Team, indicating that advanced AI models have crossed a “meaningful threshold” in developing cyber capabilities.
• Specifically, models like GLM-5.3 and Claude Mythos Preview were able to achieve “full control flow hijacks” in 4% and 6% of trials, respectively, on an “internal Binary Exploitation benchmark.”
• This represents a significant and concerning progression, as earlier models such as Claude Opus 4.6 and GLM-5.2 did not succeed in any of these binary exploitation tasks.