UK AISI finds Claude Mythos Preview can autonomously take over a simulated corporate network
UK AISI found Claude Mythos Preview completed a 32-step corporate-network takeover in three of ten attempts; Anthropic later verified practical downstream impact through large-scale real-world vulnerability discovery.
0 comments · 0 votes
Sign in to join the discussion →
No comments yet. Start the discussion.
Why it moved the index
Independent testing demonstrated end-to-end autonomous compromise across a long simulated attack chain, while later real-world vulnerability discovery verifies practical impact. Magnitude is limited because the range was weakly defended and simulated.
Assessment history
-
R1
Toward 55 · confidence 90
New historical model-specific autonomy result paired with separately dated practical cyber-impact evidence.
11 Aug 2026
Share this page
-
DoomBench assesses “UK AISI finds Claude Mythos Preview can autonomously take over a simulated corporate network” as evidence moving toward doom, with magnitude 55 and confidence 90 out of 100 in the autonomy and agency category.
-
The DoomBench assessment of “UK AISI finds Claude Mythos Preview can autonomously take over a simulated corporate network” is based on reporting from UK AI Security Institute and records the editorial rationale, source quality,...
-
DoomBench summarizes “UK AISI finds Claude Mythos Preview can autonomously take over a simulated corporate network” as follows: UK AISI found Claude Mythos Preview completed a 32-step corporate-network takeover in three of ten attempts;...
https://www.doombench.com/news/uk-aisi-finds-claude-mythos-preview-can-autonomously-take-over-a-simulated-corporate-network-2026-04-13