Skip to content
Apart Research
Apart Research
Contact us
Sprint projectMay 25, 2026Mumbai

SpecTrojan: Adversarial Specification Validation via Evil Twin Synthesis

Ojas Marathe · Team Spectacular

Submitted to The Secure Program Synthesis Hackathon. Sprint projects are early-stage work by participants, not Apart Research publications.

Read the report

Report: SpecTrojan: Adversarial Specification Validation via Evil Twin Synthesis

Open in a new tab (opens in new tab)
Recording (opens in new tab)Code (opens in new tab)More on youtu.be (opens in new tab)
Share

SpecTrojan is a spec-validation tool that inverts the traditional input-space search: given a candidate specification, an attacker LLM synthesizes an "Evil Twin" — an alternative implementation that satisfies the spec yet diverges from the reference on intent-bearing inputs. A successful twin is an artifact-level proof that the spec is too weak. On a Bio Honeypot (an LLM-written spec for a sequence-screening predicate), SpecTrojan synthesized 16 evil twins in 60 seconds, including a trivial return True that passes verification while admitting every threat-bearing input. The system also includes an attack-to-strengthening loop that auto-proposes and mechanically verifies spec repairs, and metamorphic spec testing that catches syntax-overfit specs.

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. Evil twins to show naivety is powerful. Would be useful to test fully on a bigger set, and focus more on showing robustness.

  2. The reframe is novel: search the implementation space and synthesize a whole spec-satisfying alternative, so a surviving twin is an artifact-level proof that the specification is too weak. The system is broad and well-built: a twin synthesizer, two baselines, metamorphic testing, and a strengthening loop whose verifier catches every broken LLM repair while a hand-crafted control passes, with an unusually honest robustness section. Caveats. The headline target is a constructed honeypot, and twin counts are stochastic (sixteen one run, zero the next), so foreground the stable binary signal, whether any twin is found, over the counts. A twin only satisfies the spec across eighty Hypothesis samples, so satisfaction is probabilistic. The capability matrix is rhetorically circular, its rows defined as what the method does, so drop the strictly-subsumes claim. And the central comparison figure has a truncated caption with several mangled identifiers, worth a cleanup pass.

    Read full reviewShow less

Cite this project

@misc{marathe2026spectrojan,
  title = {{SpecTrojan: Adversarial Specification Validation via Evil Twin Synthesis}},
  author = {Ojas Marathe},
  year = {2026},
  month = may,
  note = {Submitted to The Secure Program Synthesis Hackathon, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/spectrojan-adversarial-specification-validation-via-evil-twin-synthesis-qm6i}},
  url = {https://apartresearch.com/sprints/projects/spectrojan-adversarial-specification-validation-via-evil-twin-synthesis-qm6i}
}

More from The Secure Program Synthesis Hackathon

All 61 entriesBrowse all projects

Build something like this at the next Sprint

AI Collusion Research Sprint · Oct 23 - 25, 2026