Skip to content
Apart Research
Apart Research
Contact us
Sprint projectMay 24, 2026NOIDA

LLM-Assisted Ambiguity Detection in Regulatory Specifications

Kushagra Sharan · Team kshgrshrn

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

Read the report

Report: LLM-Assisted Ambiguity Detection in Regulatory Specifications

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

Regulatory software has a specification problem. Rules like "late filing attracts a penalty of Rs. 50 per day" are written for accountants and lawyers, not for engineers who need to know when the clock starts, whether there is a cap, and what happens when a supplier files late. The hard part is not implementing the rule once you understand it. The hard part is figuring out what the rule actually requires before you write any code. This project builds a lightweight pipeline that does one thing: reads a natural-language regulatory requirement and asks whether it is specific enough to implement. Three sequential LLM calls handle formalization, auditing, and scoring. The first rewrites the requirement as explicit bullet-point conditions. The second attacks the formalized version for edge cases and missing definitions. The third returns a structured JSON score and verdict. I ran it on seven GST-style compliance rules using Gemini. All seven came back underspecified, with repeated gaps around temporal boundaries, actor responsibilities, evidence requirements, and exception handling. None of that is surprising once you see it laid out, but having a system that surfaces those gaps before implementation starts is the point. The connection to secure synthesis is direct. A formally verified implementation of an incomplete specification is not safe, it is precisely wrong. The ambiguity finder sits one step before the formal methods pipeline and flags the questions that need answers first. The codebase is a single Python file with no external dependencies beyond the LLM API. A demo mode runs without any API key for reproducibility. The full Gemini output is committed under results/output.json.

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. This is a good project exploration in an interesting space. Regulatory software is, indeed, an important use-case.

  2. The security framing that a faithful implementation of an incomplete rule can let fraudulent claims pass is a good insight; the README is honest about scope and limits.

    The main gap is in evaluation: all seven sampled requirements scored "underspecified" with no negative controls and no labeled ground truth, so the difference between <regulatory text is systematically underspecified> and <the prompt always says underspecified> is unclear.

    A good next step is to add a handful of well-specified requirements as negative controls (so "adequate" can be returned) plus a few (human) expert-labeled cases to check that flagged ambiguities are real, since edge-case lists are a good deliverable.

Cite this project

@misc{sharan2026llmassisted,
  title = {{LLM-Assisted Ambiguity Detection in Regulatory Specifications}},
  author = {Kushagra Sharan},
  year = {2026},
  month = may,
  note = {Submitted to The Secure Program Synthesis Hackathon, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/llmassisted-ambiguity-detection-in-regulatory-specifications-k7v0}},
  url = {https://apartresearch.com/sprints/projects/llmassisted-ambiguity-detection-in-regulatory-specifications-k7v0}
}

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