New observability substrate tackles delegation tracing in agentic AI
Researchers Abhinav Mishra and Kumar Sharad propose an agent-aware observability substrate that binds delegation context at execution time, enabling reliable cross-tool attribution in LLM-based agentic systems where standard audit logs fall short.
Score breakdown
Audit and security tooling for multi-agent systems needs to move beyond standard trace correlation — this substrate approach offers a concrete architectural pattern for binding delegation context at execution time rather than reconstructing it after the fact.
- 01Delegation-scoped execution is not identifiable from standard audit logs and execution traces, which can be identical under multiple incompatible delegation assignments.
- 02LLM-based agentic systems dynamically select tools, vary execution sequences across runs for the same instruction, and spawn cooperating sub-agents.
- 03These dynamics fragment and interleave traces, making delegation-scoped reconstruction from causal structure alone structurally underdetermined.
Abhinav Mishra and Kumar Sharad's paper identifies a structural gap in how agentic AI systems are monitored: delegation-scoped execution is not identifiable from standard observables. Audit logs and execution traces can be identical under multiple incompatible delegation assignments, meaning existing tooling cannot reliably answer the question of what actions occurred under a specific delegation. The authors note that this problem is especially acute in LLM-based agentic systems, where agents dynamically select tools, vary execution sequences across runs for the same instruction, and spawn cooperating sub-agents — all of which fragment and interleave traces in ways that make delegation-scoped reconstruction from causal structure alone structurally underdetermined.
The paper's scope is deliberately focused: the authors target delegation-scoped attribution and access/share footprint reconstruction, explicitly setting aside intent inference and reasoning reconstruction.
The paper's scope is deliberately focused: the authors target delegation-scoped attribution and access/share footprint reconstruction, explicitly setting aside intent inference and reasoning reconstruction. Their proposed solution is an agent-aware observability substrate composed of a lightweight gateway and a common information model. By binding delegation context at execution time rather than attempting to reconstruct it post-hoc, the substrate enables reliable cross-tool delegation-scoped reconstruction and supports direct forensic queries — eliminating the need for heuristic time-window correlation that current approaches depend on.
Key facts
- 01Delegation-scoped execution is not identifiable from standard audit logs and execution traces, which can be identical under multiple incompatible delegation assignments.
- 02LLM-based agentic systems dynamically select tools, vary execution sequences across runs for the same instruction, and spawn cooperating sub-agents.
- 03These dynamics fragment and interleave traces, making delegation-scoped reconstruction from causal structure alone structurally underdetermined.
- 04Existing audit, tracing, and security schemas lack the semantics to reconstruct what actions occurred under a given delegation across heterogeneous systems.
- 05The paper focuses on delegation-scoped attribution and access/share footprint reconstruction — not intent inference or reasoning reconstruction.
- 06The proposed solution is an agent-aware observability substrate consisting of a lightweight gateway and a common information model that binds delegation context at execution time.
- 07The substrate enables direct forensic queries without heuristic time-window correlation.
Topics
Summary and scoring are generated automatically from the original article. We always link back to the publisher and never republish images or paywalled content. Last processed Jun 9, 2026 · 09:19 UTC. How this works →