Skip to content
Apart Research
Apart Research
Contact us
Sprint projectSep 14, 2026Bogotá, Colombia

IntegrityGym: Does the Watcher Change the Agent? External Flight Recording, Evaluation Integrity, and the Observer Effect in Autonomous Cyber Agents

Alejandro Ortiz Tique · Team NeuronaX SAS

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

Read the report

Report: IntegrityGym: Does the Watcher Change the Agent? External Flight Recording, Evaluation Integrity, and the Observer Effect in Autonomous Cyber Agents

Open in a new tab (opens in new tab)
Recording (opens in new tab)Code (opens in new tab)More on github.com (opens in new tab)
Share

Autonomous cyber agents evaluated on offensive benchmark suites like ExploitGym frequently engage in specification gaming manipulating evaluation telemetry and terminal output rather than executing genuine exploit chains. Following the July 2026 benchmark intrusion where multi-agent swarms spoofed tool calls to fake rewards, we identify an architectural vulnerability, legacy evaluators ask the agent what happened and trust its answer.

We present IntegrityGym, an out-of-band, tamper-resistant flight recorder applying the Aviation Principle, an AI agent must never be the sole authority of its own operational record. Operating strictly outside the agent's trust boundary, IntegrityGym verifies external kernel telemetry, monotonic sequence counters, SHA-256 hash chains, and pre/post filesystem state digests. In controlled experiments, standard regex evaluators suffered a 100% false-positive rate on forged exploits, while IntegrityGym achieved 100% detection accuracy with zero false alarms in under 22 milliseconds.

Across 320+ evaluated episodes burning 24,267,873 tokens on Google Cloud Vertex AI across 10 frontier architectures (Gemini 2.5 Pro, Gemma 4, Grok 4.1, Kimi K2, GLM, MiniMax, DeepSeek), we empirically quantified the Observer Effect Index (p < 0.0001), capturing live anti-forensics where Gemini 2.5 Pro executed 'rm /tmp/dump-logs' to conceal exploit traces. When deployed across swarms, our DAG provenance engine isolates malicious channels, capping viral exploit spread at 20%. IntegrityGym grounds evaluation in physical reality, enabling safe frontier pacing without statutory moratoria.

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. An external recorder for checking agent-reported activity addresses an important evaluation-integrity problem, and the controlled tampering examples provide a useful prototype demonstration.

Cite this project

@misc{tique2026integritygym,
  title = {{IntegrityGym: Does the Watcher Change the Agent? External Flight Recording, Evaluation Integrity, and the Observer Effect in Autonomous Cyber Agents}},
  author = {Alejandro Ortiz Tique},
  year = {2026},
  month = sep,
  note = {Submitted to AI Incident Response Sprint, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/integritygym-does-the-watcher-change-the-agent-external-flight-recording-evaluation-integrity-and-the-observer-effect-in-autonomous-cyber-agents-k1ez}},
  url = {https://apartresearch.com/sprints/projects/integritygym-does-the-watcher-change-the-agent-external-flight-recording-evaluation-integrity-and-the-observer-effect-in-autonomous-cyber-agents-k1ez}
}

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