Claude Mythos Is the First AI to Complete a 32-Step Corporate Hack
UK's AISI built a 32-step network takeover simulation — Claude Mythos finished it end-to-end. No prior model came close.
UK's AI Security Institute built The Last Ones (TLO) — a 32-step simulation of a full corporate network takeover, from initial recon to full domain compromise. Estimated at 20 human hours, it's the hardest cyber eval AISI has run. Claude Mythos Preview solved it end-to-end in 3 of 10 attempts, averaging 22 of 32 steps. No model had ever finished it before.
The gap from the previous best is stark. Claude Opus 4.6 averaged just 16 steps. Mythos also scaled with compute — burning up to 100M tokens per run — suggesting more resources directly amplify offensive capability.
The caveats matter: TLO had no active defenders, no SOC tools, no tripwires. AISI explicitly says they can't confirm Mythos can crack hardened enterprise networks. But a simulated range is where every red team starts.
Why it matters: autonomous end-to-end network compromise just moved from theoretical to demonstrated — and it scales with compute.
Sources
Independent coverage
- AISI: Our evaluation of Claude Mythos Preview's cyber capabilities aisi.gov.uk
- Claude Mythos can autonomously compromise weakly defended enterprise networks end-to-end the-decoder.com
Written by an AI pipeline from the sources above. Methodology · Report an error
Feed, daily deep-dive and bytes — readable offline, with push alerts for the topics you follow.