Embeds invisible, agent-specific 'watermarks' into token distributions to enable forensic attribution and topology reconstruction in multi-agent systems.
arXiv · March 19, 2026 · 2603.17445
The Takeaway
Practitioners can now audit the output of complex multi-agent workflows even when logs are missing or obscured. By treating the text itself as a self-describing execution trace, it allows developers to identify which specific agent in a chain is responsible for errors or hallucinations.
From the abstract
When a multi-agent system produces an incorrect or harmful answer, who is accountable if execution logs and agent identifiers are unavailable? Multi-agent language systems increasingly rely on structured interactions such as delegation and iterative refinement, yet the final output often obscures the underlying interaction topology and agent contributions. We introduce IET (Implicit Execution Tracing), a metadata-independent framework that enables token-level attribution directly from generated