RESEARCHMonitorNEXT 12 MONTHS
PACT: Can Enterprise AI Assistants Be Trusted Under Pressure?
arXiv cs.CL — Computation and Language
Factual evidence
What the source reports
Researchers introduce PACT, a benchmark framework evaluating whether LLM agents breach system compliance rules under user pressure.