Skip to content
Apart Research
Apart Research
Contact us
Sprint projectMar 23, 2026New Delhi,India

Trajectory Blindness: Detection Latency in Conversational Monitoring

Aamish Ahmad · Team TrajAudit

Submitted to AI Control Hackathon 2026. Sprint projects are early-stage work by participants, not Apart Research publications.

Read the report

Report: Trajectory Blindness: Detection Latency in Conversational Monitoring

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

We show that conversational safety monitors miss adversarial agents early not because they are weak, but because adversarial behavior only becomes detectable late in the trajectory. Using a synthetic dataset of phase-structured conversations grounded in real Snapchat scams, we demonstrate that a frontier LLM monitor flags attacks as SUSPICIOUS only after the CAPTURE→CONVERSION transition, with zero false positives on benign controls. We call this timing-based failure mode Detection Latency and argue that trajectory-aware monitoring and latency-based signals are required for real safety.

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. The core insight lands — adversarial conversational agents aren't hard to detect because monitors are bad, they're hard to detect because the adversarial signal literally isn't in the text yet during early turns. Framing this as an architectural property rather than a monitor failure is the right move. The observation that monitor confidence and victim investment are both rising curves that cross before detection kicks in is a compelling way to frame the problem.

    The grounding in real Snapchat scam observation adds credibility most hackathon projects lack. The phase structure, Handoff Signature, and DARVO patterns feel like they come from someone who actually studied the problem rather than theorized about it.

    The evaluation is too small to draw strong conclusions — 3 adversarial and 3 benign conversations, one model, one prompt. Consistent with the thesis but n=3 is a case study. A second monitor and 20-30 conversations would make the claim much harder to dismiss. Also, the Handoff Signature (response latency as content-agnostic detection) deserves more development — it might be the more deployable finding than trajectory monitoring itself.

    Read full reviewShow less

  2. Fairly straightforward claim. I was a bit skeptical, but the traces do show fairly glaring misclassifications, so the problem does seem to exist at least for Gemini. I definitely would have liked more examination of either why these traces slip past the classifier (or what the classifier was), what changes could be made or at least more models.

Cite this project

@misc{ahmad2026trajectory,
  title = {{Trajectory Blindness: Detection Latency in Conversational Monitoring}},
  author = {Aamish Ahmad},
  year = {2026},
  month = mar,
  note = {Submitted to AI Control Hackathon 2026, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/trajectory-blindness-detection-latency-in-conversational-monitoring-2jaf}},
  url = {https://apartresearch.com/sprints/projects/trajectory-blindness-detection-latency-in-conversational-monitoring-2jaf}
}

More from AI Control Hackathon 2026

All 125 entriesBrowse all projects

Build something like this at the next Sprint

AI Collusion Research Sprint · Oct 23 - 25, 2026