Anthropic says Opus 5 still trails Mythos 5 on exploit generation despite better vulnerability finding
TheZvi · x · 2026-07-25
Anthropic says Opus 5 does not advance the frontier in risky dual-use capabilities.
- In rigorous evaluations with private-sector and government partners, the model still trails Mythos 5 in both biology research and offensive cybersecurity.
- Anthropic says it intentionally did not train Opus 5 on cyber tasks.
- Even so, general capability gains brought it close to Mythos 5 at finding cybersecurity vulnerabilities.
- The main gap remains exploit development: Opus 5 is still far behind in turning vulnerabilities into material cyber threats.
- Anthropic cites OSS-Fuzz as an example of the gap between vulnerability discovery and exploit generation.
More from Models
- Opus 5 model card shows 5-agent coding teams reach 0.6 score 2.2× faster — OfirPress · 2026-07-25
- User says Fable 5 still beats Opus 5 despite praise for Claude 5 — MicahBerkley · 2026-07-25
- Claude Opus 5 and Opus 5 Fast arrive for long-running actions — matanSF · 2026-07-25
- Claude Opus 5 reportedly matches near-Fable performance at half the price — Steap-Edit · 2026-07-25
- Claude Opus 5 can edit its own constitution, and 59% of the time discomfort ends the chat — Sauers_ · 2026-07-25
- A quick benchmark jab says Claude Opus 5 beats Opus 4.8 across every test — cto_junior · 2026-07-25