Pratibha Revankar
1 indexed paper
Recent (6 mo)
1With code
0Influential cites
0Benchmarked
0Publications per year
126
Top categories
Crypto×1AI×1
Frequent co-authors
Research Timeline
2026
Caught in the Act(ivation): Toward Pre-Output and Multi-Turn Detection of Credential Exfiltration by LLM Agents
This paper proposes a multi-layered defense strategy combining pre-output monitoring, calibrated canary detection, and cumulative information-flow tracking to prevent LLM agents from exfiltrating sensitive credentials.
Highlighted terms show continued research focus across papers