A joint report from the Alan Turing Institute's CETaS and Georgetown's CSET assesses autonomous cyber defence, focusing on reinforcement learning (RL) agents trained in cyber gyms such as CAGE. It sets out technical challenges (combinatorial action spaces, reward design, transferability, securing the defender agents) and policy challenges (human-machine teaming, testing, liability), and recommends investment in gyms, test ranges, competitions and authorisation thresholds.
Why it matters
It records how autonomous defence was framed before LLM agents entered the field, including the open question of when defensive agents may act without human approval.
Key facts
As stated in the sources, with where to find them.
- Authors are Andrew Lohn, Anna Knack, Ant Burke and Krystal Jackson; the report is dated June 2023.Title page
- The report scopes autonomous cyber defence to RL-based agents and lists recommendations including scaling gyms, building test ranges, hosting competitions and determining thresholds for authorising autonomous defence agents.Section 4 'Conclusions and Recommendations'
Findings that cite this record
No tracked finding cites this record yet.
Key questions this bears on
- Can prompt injection against AI agents be reliably defended?Not reliably. Bounding what untrusted input can trigger helps, but trust boundaries must survive delegation and tool execution.
Sources
Related records
Oct 27, 2026
Oct 13, 2026
Oct 12, 2026
Sep 30, 2026
Sep 28, 2026
Sep 26, 2026