Engine-managed context branching plus prospective label checks confine taint in LLM agents, cutting exfiltration to 0–7% ASR while recovering much of the utility pure taint tracking forfeits.
Schwartz, Thanassis Avgerinos, and David Brumley
1 Pith paper cite this work, alongside 702 external citations. Polarity classification is still indexing.
1
Pith paper citing it
702
external citations · OpenAlex
fields
cs.CR 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Agentic Permissions Policy Algebra for Taint Confinement in LLM Agents
Engine-managed context branching plus prospective label checks confine taint in LLM agents, cutting exfiltration to 0–7% ASR while recovering much of the utility pure taint tracking forfeits.