Skip to content

Back to the catalog

simonwillison.netwriteup

Quoting Anthropic Frontier Red Team

A quoted excerpt from Anthropic's Frontier Red Team report on GLM-5.3 finds that, across 100 randomly selected tasks from an internal Binary Exploitation benchmark, GLM-5.3 achieved full control-flow hijacks in 4% of trials and Claude Mythos Preview in 6%, while earlier models (Claude Opus 4.6, GLM-5.2) succeeded in none. Anthropic frames this as a meaningful threshold crossed in advanced cyber capability, part of ongoing frontier-model dangerous-capability evaluation.

TL;DR Anthropic's Frontier Red Team finds GLM-5.3 and Claude Mythos Preview newly capable of full binary-exploitation control-flow hijacks, unlike prior-generation models.
Open original ↗
Also seen from another angle🛡️ Security →

Organisations named here

  • Anthropic
  • Anthropic Frontier Red Team
  • Simon Willison

Technologies named here

  • GLM-5.3
  • Claude Mythos Preview
  • Claude Opus 4.6
  • GLM-5.2

Related in ai-safety