Skip to content
Apart Research
Apart Research
Contact us
Sprint projectSep 14, 2026San Francisco

Ryan Junejo

Ryan Junejo · Team Crashlabs

Submitted to AI Incident Response Sprint. Sprint projects are early-stage work by participants, not Apart Research publications.

Read the report

Report: Ryan Junejo

Open in a new tab (opens in new tab)
Code (opens in new tab)
Share

Receipt possession is not event attribution

EvidenceGraph is a forensic tool that rebuilds what an AI agent did during an incident by cross-checking agent transcripts against independent platform records, such as a registry’s write log. This sprint project tested one narrow failure in that reconstruction: a transcript event carrying a genuine receipt copied from a different event.

The failure was real. Under the old rule, copying a receipt with its token into another transcript made the tool attribute one of twelve real registry writes to the wrong event, with full confidence. The fix makes the missing assumption explicit. Matching tokens now stay ambiguous unless the case declares that the recorder is authentic or that tokens could not have been copied. Analyzer upgrades also invalidate old conclusions until the case is recomputed, so stale attributions cannot survive the safer rule.

Twelve paired evidence conditions across three seeds behaved as specified in all 36 checks, including a deliberately false declaration that brings the wrong attribution back. The cost is visible too: on clean evidence with unknown token exclusivity, the tool withholds all twelve correct attributions. An evidence-collection checklist tells investigators what to gather before making stronger claims.

Everything ran on synthetic logs with no model, external target or credential. The results are development checks on one scripted workflow, not estimates of accuracy on real incidents or of usefulness to investigators.

Reviews

Judging this Sprint?

Review this project

Your public critique appears on this page without your name. Your private critique is not published; only the Apart team reads it. If you agree below, we share your review with grantmaking.ai (opens in new tab) and the Transformative AI Fund so strong projects can be funded.

Not shown on this page.

Shown on this page, without your name.

Only the Apart team reads this, and funders if you agree below.

Share my name publicly on grantmaking.ai *
Share my private critique with funders *

How much would this matter for AI safety if it worked? How innovative is it? For scores of 4-5: is this actually new to the field, or replicating recent work?

Scoring guide
  1. 1Negligible. No clear problem addressed, or no meaningful novelty.
  2. 2Limited. Addresses a real problem but with a generic or well-trodden approach. Incremental at best.
  3. 3Moderate. Clear problem with a reasonable approach; some novelty in framing or method beyond routine application of existing tools.
  4. 4Significant. Important problem with an original approach, or identifies a neglected problem area. A valuable contribution others could build on.
  5. 5Exceptional. Tackles a critical AI safety problem with a genuinely novel approach, or opens a new research direction. Clear theory of change. You'd be excited to share this with researchers in the area.

How sound are methodology, implementation, and findings?

Scoring guide
  1. 1Seriously flawed. Methodology broken, results uninterpretable, or implementation doesn't work.
  2. 2Weak. Approach has significant gaps: missing validation, flawed experimental design, or incomplete implementation.
  3. 3Competent. Technically solid given the short duration. Methodology makes sense, results are interpretable, limitations acknowledged, work builds toward clear conclusions.
  4. 4Strong. Thorough methodology with convincing validation. Results clearly support conclusions. Immediately useful for future work.
  5. 5Exceptional. Ambitious scope executed rigorously. Surprising findings, novel methods, or unusually robust validation.

How clearly are work, findings, and impact potential communicated?

Scoring guide
  1. 1Incomprehensible. Cannot determine what the project is actually claiming or doing.
  2. 2Hard to follow. Key information buried, missing, or diluted by excessive length. Significant effort to extract main points.
  3. 3Clear enough. Can understand the problem, approach, and results without undue effort. Core content clearly present: problem, method, findings, limitations.
  4. 4Well presented. Easy to follow, well-structured, appropriate level of detail. Target audience would get it quickly.
  5. 5Exceptionally clear. A pleasure to read. Complex ideas made accessible. Could serve as a model for how to present this type of work.

  1. Nice story: the team put their own tool to the test, found a real weak spot (it could be tricked into blaming the wrong event), and fixed it properly. Small in scope, but a genuine, well-handled catch.

  2. This forensics project has a good approach. As AI incidents can produce large amounts of evidence, being able to distinguish whether an action happened from which recorded event actually caused it is important. I would like to see this tested on a larger simulated incident with more realistic agent behaviour and evidence, to see whether the attribution approach remains useful outside the small synthetic workflow used here.

Cite this project

@misc{junejo2026ryan,
  title = {{Ryan Junejo}},
  author = {Ryan Junejo},
  year = {2026},
  month = sep,
  note = {Submitted to AI Incident Response Sprint, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/ryan-junejo-4omn}},
  url = {https://apartresearch.com/sprints/projects/ryan-junejo-4omn}
}

More from AI Incident Response Sprint

All 233 entriesBrowse all projects

Build something like this at the next Sprint

AI Collusion Research Sprint · Oct 23 - 25, 2026