Research team proposes evidence tracing for LLM agent auditing

Academic paper introduces methods to trace and verify LLM agent decision-making through execution provenance, addressing audit and compliance needs for autonomous AI systems in high-stakes domains.