The DOI matches its Crossref record; 1 of the 1 references it cites were found in a registry or an index.
Creating Characteristically Auditable Agentic AI Systems
Open accessAbstract
Auditability -the ability to reliably record, verify, and analyze agent behavior -is a key requirement for trustworthy multi-agent AI systems.This paper formalizes auditability with a set of eight axioms (Integrity, Coverage, Temporal Coherence, Verifiability, Accessibility, Resource Proportionality, Privacy Compatibility, and Governance Alignment) and extends the theory to incorporate liveness properties (ensuring eventual auditability under fault tolerance and agent failures) and adversarial resilience (game-theoretic resistance to strategic log manipulation).We situate this framework in the context of recent research on accountability, transparency, and traceability in distributed AI, and we identify design patterns that operationalize these auditability axioms in system architecture and agent behavior (including partial observability, asynchronous communication, and incentive constraints).All propositions are stated with formal rigor and proved inline.The resulting formalism provides a foundation for building multi-agent systems that are not only auditable by design but also robust against faults and adversaries, aligning with emerging governance requirements for AI.
Overview
The claims and statistics of this paper haven’t been checked yet.
Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
The cited source says what the claim attributes to it.
In the citing paper
“Externally, they support debugging, compliance auditing, and postincident analysis [Phiri, 2025].”
From Externalization in LLM Agents: A Unified Review of Memory, Skills, Protocols and Harness Engineering (2026) · cites this paper as Phiri, 2025
The cited source says what the claim attributes to it.
In the citing paper
“when every action is documented transparently, investigators can trace outcomes back through the sequence of agent decisions Phiri (59).”
From Ethical issues in multi-agent AI systems for healthcare: a narrative review (2026) · cites this paper as [59]
The check wasn’t confident either way, so it gives no verdict — never disputed.
In the citing paper
“Direct LLM output is brittle when used alone.Correctness comes from verifiers, repair loops, and explicit checks around the generator [75], [89], [90].”
From Large language models for agentic NetOps and AIOps: Architectures, evaluation, and safety (2027) · cites this paper as [90]
Experimental — accuracy not yet measuredThe check wasn’t confident either way, so it gives no verdict — never disputed.
In the citing paper
“Automatic Configuration Repair is illustrative: it frames misconfiguration handling as localise-fix-validate, which matches how operators actually de-risk changes during incidents [90].”
From Large language models for agentic NetOps and AIOps: Architectures, evaluation, and safety (2027) · cites this paper as [90]
Experimental — accuracy not yet measured
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Citations
Cites
Accountability in artificial intelligence: what it is and how it works
Novelli, C., Taddeo, M., & Floridi, L. (2024) · AI & SOCIETY · doi:10.1007/s00146-023-01635-y · Cited by 495 · cited 1×
“Novelli et al. (2024) identify accountability with answerability and consequences, outlining conditions that any accountable AI must satisfy [12].”
Research works printed but not linked: 14
- 10.1145/2723372.2723711 · DOI
- Matteo Baldoni Cristina Baroglio Roberto Micalizio and Stefano Tedeschi. 2023. Accountability in Multi‐Agent Organizations: From Conceptual Design to Agent Programming. Autonomous Agents and Multi‐Agent Systems 37 1 (2023) 7:1–7:35. 10.1007/s10458-022-09590-6 · DOI
- Ben C. Cheong. 2024. Transparency and Accountability in AI Systems: Safeguarding Well-Being in the Age of Algorithmic Decision‐Making. Frontiers in Human Dynamics 6 (2024) 1421273. 10.3389/fhumd.2024.1421273 · DOI · Paper page
- Cynthia Dwork Nancy Lynch and Larry Stockmeyer. 1988. Consensus in the Presence of Partial Synchrony. J. ACM 35 2 (1988) 288–323. 10.1145/42282.42283 · DOI
- 10.1145/3442188.3445905 · DOI
- Leslie Lamport. 1978. Time Clocks and the Ordering of Events in a Distributed System. Commun. ACM 21 7 (1978) 558–565. 10.1145/359545.359563 · DOI · Paper page
- Leslie Lamport. 1998. The Part‐Time Parliament. ACM Transactions on Computer Systems 16 2 (1998) 133–169. 10.1145/279227.279229 · DOI
- 10.1145/800217.806614 · DOI
- JongHyup Lee. 2022. DPTracer: A Log‐Driven Accountability System for Distributed Data Provision Networks. Applied Sciences 12 23 (2022) 12129. 10.3390/app122312129 · DOI
- Marçal Mora‐Cantallops Salvador Sánchez‐Alonso Elena García‐Barriocanal and Miguel‐Ángel Sicilia. 2023. Traceability for Trustworthy AI: A Review of Models and Tools. Big Data and Cognitive Computing 7 1 (2023) 5. 10.3390/bdcc7010005 · DOI
- Anudit Nagar, Cuong Tran, and Ferdinando Fioretto. 2021. Privacy‐Preserving and Accountable Multi‐Agent Learning. In Proc. 20th International Conference on Autonomous Agents and Multi‐Agent Systems (AAMAS 2021). International Foundation for Autonomous Agents and Multiagent Systems, London, UK, 1605–1607.
- OpenTelemetry Project. 2024. Maintainer Benchmark and Security Audit Report. Technical Report. Cloud Native Computing Foundation. https://opentelemetry.io.
- David L. Parnas. 1972. On the Criteria To Be Used in Decomposing Systems into Modules. Commun. ACM 15 12 (1972) 1053–1058. 10.1145/361598.361623 · DOI
- Stelios Triantafyllou, Adish Singla, and Goran Radanovic. 2021. On Blame Attribution for Accountable Multi‐Agent Sequential Decision Making. In Advances in Neural Information Processing Systems 34 (NeurIPS 2021). NeurIPS, Virtual Conference, 1–12. Paper 7109; arXiv:https://arXiv.org/abs/2107.11927. · arXiv
Cited by
Sorted by citations
Externalization in LLM Agents: A Unified Review of Memory, Skills, Protocols and Harness Engineering
Zhou, C., Chai, H., Chen, W., Guo, Z., Shan, R., Song, Y., Xu, T., Yang, Y., Yu, A., Zhang, W., et al. (2026) · arXiv · doi:10.48550/arxiv.2604.08224 · Cited by 1
Ethical issues in multi-agent AI systems for healthcare: a narrative review
Xie, Z., Wang, H., Dai, L., Wang, Z., Song, H., & Qian, J. (2026) · Frontiers in Public Health · doi:10.3389/fpubh.2026.1792627 · Cited by 11
Large language models for agentic NetOps and AIOps: Architectures, evaluation, and safety
Bilal, M., Crowcroft, J., Wang, R., Xu, X., & Dustdar, S. (2027) · Computer Science Review · doi:10.1016/j.cosrev.2026.101075 · Cited by 10
Metadata
- Authors
- Phiri, C. C.
- Affiliations
- JPMorgan Chase & Co (United States) (US)
- Corresponding author
- Phiri, Charles Chimwemwe
- Venue
- Proceedings of the Intelligent Robotics FAIR 2025
- Publisher
- ACM
- Published
- 2025-06-23
- Pages
- 1-14
- OpenAlex
- W7084133848
- Cited by
- 7
- References
- 1
- Citations by year
- 2026: 7
- FWCI
- 23.43
- Type
- conference paper
- Language
- en
- Open access
- hybrid
- doi.org
- Licence
- cc-by
- Locations
- Intelligent Robotics FAIR (publishedVersion, cc-by)
- Topics
- Physical Sciences › Computer Science › Computer Networks and Communications › Access Control and Trust
- Physical Sciences › Computer Science › Artificial Intelligence › Multi-Agent Systems and Negotiation
- Physical Sciences › Computer Science › Artificial Intelligence › Ethics and Social Impacts of AI
- Keywords
- auditability, agentic AI, multi-agent systems, liveness properties, fault tolerance, adversarial resilience, traceability, transparency, accountability, AI governance, partial observability, asynchronous communication, incentive constraints, trustworthy AI
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
