🧰 Claude Security

The items that name it 5

• Frontier models found the vulnerabilities. Only the attacker found the chains.
  article2026-10-07It is written here as Claude Security (running Mythos).

  AI-driven vulnerability discovery tool that reads and reasons about source code (white-box, code only), with extended thinking; tested Sep 18 against TaintedPort, found one more critical-severity vulnerability than Evo COS (10 vs 9), precision 90.2%, caught source-visible logic and cryptographic fla

• Claude Security
  tool2026-07-23

  Scans codebases (via connected GitHub repositories or directories) using Claude Mythos 5.1 to find vulnerabilities, validates findings through an adversarial self-verification pass, and suggests patches for human review and approval; available in public beta for Claude Enterprise; can push findings

• Defending Code Reference Harness
  repo2026-06-05

  Anthropic's managed/hosted product that scans repositories for vulnerabilities, applies a multi-stage verification pipeline to reduce false positives, and manages findings through triage, fix validation, and rapid fix generation

• Project Glasswing: Initial Update (Anthropic)
  article2026-05-27

  Tool in public beta for Claude Enterprise customers that scans codebases for vulnerabilities and generates proposed fixes; powered by Claude Opus 4.7, used to patch over 2,100 vulnerabilities in three weeks since launch.

• dfbench — depthfirst
  article

  Used to run the Mythos 5* entry in the dfbench results (marked with an asterisk).

In every workspace you can read 7