When Only the Final Text Survives: Implicit Execution Tracing for Multi-Agent Auditing
When a multi-agent system produces an incorrect or harmful answer, who is accountable if execution logs and agent identifiers are unavailable? In practice, generated content is often detached from its execution environment due to privacy or system boundaries, leaving the final text as the only auditable artifact. Existing attribution methods rely on full execution traces and thus become ineffective in such metadata-deprived settings. We propose Implicit Execution Tracing (IET), a provenance-by-design framework that shifts attribution from post-hoc inference to built-in instrumentation. Rather than inferring provenance after the fact, IET embeds agent-specific, key-conditioned statistical ...
| Authors | Yi Nian, Haosen Cao, Shenzhe Zhu et al. |
|---|---|
| Primary category | cs.AI |
| Published | 18 March 2026 |
| Last updated at source | 1 August 2026 |
Silytics shows public metadata from arXiv and links to the original. Information may be incomplete or out of date. Last refreshed 17 October 2026. Licenses and terms of the original apply to the work itself. Not financial, legal or professional advice.