The authors built a reference monitor around an unmodified agent rather than changing its model, prompt, tools or framework.
Kairos delays agent actions until their consequences reveal security risks
The system stages uncertain file and network effects outside an unmodified LLM agent, then approves or rejects them using the accumulated execution trace.
Academic
Kristoffer Hempel · Nils Loose · Lorenzo Cavallaro · Thomas Eisenbarth
Universität zu Lübeck · University College London · Mohamed bin Zayed University of Artificial Intelligence
Research Digest··3 min read
Hempel et al.
Why this paper
From Universität zu Lübeck and 2 others
In one line
Kairos reduces LLM agent attack success from 47% to 2.7% by monitoring every effect at the system isolation boundary.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ·No stated limitations found
- ✓Reports numbers on named benchmarks
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§