# Kainan and Hershman propose Agentic Process Evidence [Canonical reading](https://escapemeridian.com/readings/kainan-and-hershman-propose-agentic-process-evidence-11787823) Yuval Kainan and Carmit Hershman argue that AI agents’ work in software development creates an accountability gap: their reasoning, decisions, and tool calls vanish when a session ends. They identify three procedural problems: supply-chain security risk, policy compliance, and alignment between an agent’s tasks and goals; they explicitly set scale aside. Their proposal, Agentic Process Evidence (APE), is a signed, tamper-evident, machine-readable attestation that links agent sessions to their outcomes. The authors say APE enables Policy-as-Code control and allows a human in the loop when needed. They also say the approach supports an in-toto style attestation framework and is aligned with relevant regulations. The idea matters because it aims to connect an agent’s process with the outcome, addressing the accountability problem the authors describe. The abstract does not report an evaluation or explain how APE would work in practice, so its effectiveness remains unclear. Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims. ## Original source arXiv eess.SY · Article Original source date: Oct 29, 2026 (2026-10-29). Date basis: Date basis not provided. Added to Escape Meridian: 2026-11-11T10:54:49.509Z. First observed: 2026-10-30T06:42:34.298Z. Retrieved: 2026-10-31T04:01:59.358Z. - [Agentic process evidence](https://arxiv.org/abs/2610.40071) ## Limits The evidence is the arXiv abstract; it does not provide implementation details or report an evaluation of APE. The authors explicitly say they are not addressing scale. ## Open question What information would an APE attestation contain, and how would it link an agent session to its outcome? ## Source-stated Kainan and Hershman argue that agent activity creates an accountability gap because agents’ reasoning, decisions, and tool calls disappear after a session. > their reasoning, decisions, and tool calls vanish the moment the session ends - [Supporting source](https://arxiv.org/abs/2610.40071) ## Source-stated The authors identify supply-chain security, policy compliance, and task alignment as three procedural problems, excluding scale. > 3 main procedural problems (not addressing scale) - Supply-chain vector - [Supporting source](https://arxiv.org/abs/2610.40071) ## Source-stated They introduce APE as a signed, tamper-evident, machine-readable attestation linking agent sessions to outcomes. > a signed, tamper-evident, machine-readable attestation of agentic processes, linking the sessions to their outcome - [Supporting source](https://arxiv.org/abs/2610.40071) ## Source-stated The authors say APE supports Policy-as-Code control and allows human involvement when needed. > APE enables Policy-as-Code control over the agentic processes, allowing for human in the loop only when it's needed. - [Supporting source](https://arxiv.org/abs/2610.40071) ## Source and contributor context Kainan, Yuval, Hershman, Carmit · Article author These are credits stated by the source; they do not establish a profile match or authorship of every passage. ## Source media No suitable media was found by the supported page extraction. ## Terms and references Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement. ## Agentic AI AI systems that plan and carry out a series of steps toward a goal, often by using other software, rather than giving one answer. - [Definition reference](https://developers.google.com/machine-learning/glossary#agentic) ## AI agent An AI system that chooses and carries out actions toward a goal, often using software tools. - [Definition reference](https://www.ibm.com/think/topics/ai-agents) ## arXiv arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed. - [Explore this context](/entities/arxiv) - [Reference](https://arxiv.org/) - [Reference](https://info.arxiv.org/help/api/index.html) ## Editorial provenance Version 1. Publication ID: 446353db-c8c3-4697-8fb4-cb03009cc1ec. Prepared by gpt-6-luna; grounding review by gpt-6-luna. Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-11T10:54:49.509Z. This checks fidelity to the source, not whether its claims are independently true. Editorial selection: accept; reasons: material\_development. Observation: 06519ce3eb4e2e26f8368ab712091b3389bc67d64fc92a270eb2c0e20ee75294. Source text hash: c1adb2f1871339bf4c92c68ba6aeaef0222edf80db653368fffb7415f04c004c. ## Revision and source-check history ### Version 1 — 2026-11-11T10:54:49.509Z Yuval Kainan and Carmit Hershman argue that AI agents’ work in software development creates an accountability gap: their reasoning, decisions, and tool calls vanish when a session ends. They identify three procedural problems: supply-chain security risk, policy compliance, and alignment between an agent’s tasks and goals; they explicitly set scale aside. Their proposal, Agentic Process Evidence (APE), is a signed, tamper-evident, machine-readable attestation that links agent sessions to their outcomes. The authors say APE enables Policy-as-Code control and allows a human in the loop when needed. They also say the approach supports an in-toto style attestation framework and is aligned with relevant regulations. The idea matters because it aims to connect an agent’s process with the outcome, addressing the accountability problem the authors describe. The abstract does not report an evaluation or explain how APE would work in practice, so its effectiveness remains unclear. ## Original source arXiv eess.SY · Article Original source date: Oct 29, 2026 (2026-10-29). Date basis: Date basis not provided. Added to Escape Meridian: 2026-11-11T10:54:49.509Z. First observed: 2026-10-30T06:42:34.298Z. Retrieved: 2026-10-31T04:01:59.358Z. - [Agentic process evidence](https://arxiv.org/abs/2610.40071) ## Limits The evidence is the arXiv abstract; it does not provide implementation details or report an evaluation of APE. The authors explicitly say they are not addressing scale. ## Open question What information would an APE attestation contain, and how would it link an agent session to its outcome? ## Source-stated Kainan and Hershman argue that agent activity creates an accountability gap because agents’ reasoning, decisions, and tool calls disappear after a session. > their reasoning, decisions, and tool calls vanish the moment the session ends - [Supporting source](https://arxiv.org/abs/2610.40071) ## Source-stated The authors identify supply-chain security, policy compliance, and task alignment as three procedural problems, excluding scale. > 3 main procedural problems (not addressing scale) - Supply-chain vector - [Supporting source](https://arxiv.org/abs/2610.40071) ## Source-stated They introduce APE as a signed, tamper-evident, machine-readable attestation linking agent sessions to outcomes. > a signed, tamper-evident, machine-readable attestation of agentic processes, linking the sessions to their outcome - [Supporting source](https://arxiv.org/abs/2610.40071) ## Source-stated The authors say APE supports Policy-as-Code control and allows human involvement when needed. > APE enables Policy-as-Code control over the agentic processes, allowing for human in the loop only when it's needed. - [Supporting source](https://arxiv.org/abs/2610.40071) ## Source and contributor context Kainan, Yuval, Hershman, Carmit · Article author These are credits stated by the source; they do not establish a profile match or authorship of every passage. ## Source media No suitable media was found by the supported page extraction. ## Terms and references Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement. ## Agentic AI AI systems that plan and carry out a series of steps toward a goal, often by using other software, rather than giving one answer. - [Definition reference](https://developers.google.com/machine-learning/glossary#agentic) ## AI agent An AI system that chooses and carries out actions toward a goal, often using software tools. - [Definition reference](https://www.ibm.com/think/topics/ai-agents) ## arXiv arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed. - [Explore this context](/entities/arxiv) - [Reference](https://arxiv.org/) - [Reference](https://info.arxiv.org/help/api/index.html) ## Editorial provenance Version 1. Publication ID: 446353db-c8c3-4697-8fb4-cb03009cc1ec. Prepared by gpt-6-luna; grounding review by gpt-6-luna. Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-11T10:54:49.509Z. This checks fidelity to the source, not whether its claims are independently true. Editorial selection: accept; reasons: material\_development. Observation: 06519ce3eb4e2e26f8368ab712091b3389bc67d64fc92a270eb2c0e20ee75294. Source text hash: c1adb2f1871339bf4c92c68ba6aeaef0222edf80db653368fffb7415f04c004c. ### uncertain — 2026-11-11T10:54:49.509Z — anthropic/claude-opus-5.5 Publication: 446353db-c8c3-4697-8fb4-cb03009cc1ec attribution — pass: The draft correctly attributes the argument and proposal to the arXiv authors, Yuval Kainan and Carmit Hershman. It frames the claims as the authors' own rather than as established facts. > Authors:Yuval Kainan, Carmit Hershman summary — uncertain: The core content is accurate, but the summary drops the authors' central framing: agents should be treated as an insider threat, and the governance assumption that 'a human made the decision' no longer holds. That framing gives the source its force. Smallest repair: add one sentence on the insider-threat and human-decision argument. Key terms are also left unexplained for general readers: attestation, Policy-as-Code, in-toto and human in the loop. Repair: add term-library tooltips. The question is substantive and follows from the stated gap. > forcing us to address AI Agents as an insider threat date — pass: The originalPublished date, 2026-10-29, matches the v1 arXiv submission date. It is not the later feed date of 2026-10-31. > \[Submitted on 29 Oct 2026\] support — uncertain: Most claims are source-stated. However, the source says APE allows a human in the loop 'only when it's needed'. The draft and the fourth claim drop 'only', which softens a limiting condition into a general allowance. Smallest repair: restore 'only when needed'. The draft's limitation that no evaluation or implementation detail is given is supported by the abstract. > allowing for human in the loop only when it's needed media — pass: The selected media is the verified direct link to the original arXiv abstract page. It is the most direct way to access the source. ## Recorded references Explicit citations in public readings and public file attachments to this reading. Shared topics are not citations. Other artifact types need an explicit reference adapter. No public references recorded here yet. [Reference data (JSON)](/readings/1178782395c2dab5698f17e9b34533980680c98a680bb7b6513da3ec038a5c12/references.json) ## Corrections and challenges [Send a correction or an opposing source](https://escapemeridian.com/corrections/) [JSON record](https://escapemeridian.com/readings/kainan-and-hershman-propose-agentic-process-evidence-11787823.json)