Policy
Iranian Judge Proposes 'Reverse-Traceability' Test for AI Drafts
A civil court judge argues that verifying citations isn't enough—courts must trace final rulings back to every argument that could change the outcome.
Key takeaways
- An Iranian civil judge proposes a "reverse-traceability" test requiring courts to trace final rulings back to every potentially decisive argument in the original filings.
- The test addresses a structural risk: AI-generated judgments can contain accurate citations while omitting objections that could change the outcome.
- Judges would identify objections that, if correct, would defeat the ruling, then verify the final reasons address each one—not just mention keywords.
- The framework provides an operational method for the human oversight promised in England and Wales' October 2025 AI guidance.
- Amirkhani emphasizes the safeguard must work without approved technology and should not demand exhaustive judgments on immaterial points.
A new framework for judicial AI oversight
As courts worldwide experiment with artificial intelligence to draft judgments and summarize case files, a fundamental question has emerged: How can judges verify that AI-assisted rulings actually address the arguments that matter most?
Dr. Ehsan Amirkhani, presiding judge of Branch 2 of the General Civil Court of Kermanshah, Iran, has proposed a concrete answer. Writing in JURIST, he introduces a "reverse-traceability" test designed to catch a specific failure mode that citation-checking alone cannot prevent—the omission of outcome-determinative objections.
The structural risk of AI summarization
The problem Amirkhani identifies is not fabricated quotes or hallucinated case law. It's something more subtle: a judgment that contains accurate legal citations and fluent reasoning while completely bypassing an argument that could defeat its conclusion.
Consider a hypothetical possession claim where service of notice is legally required. An AI system might generate a draft that correctly discusses property law, cites genuine precedents, and grants possession—yet never addresses disputed evidence that the required notice was never served. Each step in the automation chain could appear internally consistent while the original omission persists through summarization, drafting, and review.
This structural vulnerability exists because comparing a draft only against the summary that generated it offers no safeguard against what the summary itself omitted.
Why it matters
Courts in England, Wales, and other jurisdictions are actively deploying AI tools for case summarization and document analysis. The October 2025 position statement from England and Wales emphasizes human oversight but doesn't specify how judges should verify completeness when the danger is a missing argument rather than a false one. Amirkhani's framework provides an operational method for the accountability those guidelines promise.
How reverse-traceability works
The proposed test begins not with the AI draft but with the authoritative record: pleadings, written submissions, hearing transcripts, and admitted evidence. The judge identifies objections that, if correct, would defeat an essential factual or legal step in the proposed ruling.
The test then asks a counterfactual question: If this objection were well-founded, could the order still stand on the grounds given? If not, the final judgment must address that disputed premise, apply the relevant rule, or explain why the objection fails.
This differs from citation verification in two ways. First, it tests completeness against a source independent of the draft itself. Second, it links completeness to legal consequence rather than mere mention—a keyword appearing somewhere in the text is not enough if the reasoning doesn't actually engage the point.
Amirkhani emphasizes that the method must remain proportionate. A straightforward dispute might contain one decisive objection; a complex appeal could have several. No fixed number should be imposed, and the judge—not an algorithm—must determine which arguments are potentially outcome-determinative.
Implementation and boundaries
The framework assumes judges retain access to underlying documents and can challenge automated summaries. Procurement policies should preserve traceable access to sources rather than offering polished interfaces that conceal the record.
Amirkhani acknowledges practical concerns: judges already read files, and the test shouldn't demand exhaustive judgments addressing every minor point. The standard remains materiality, assessed in light of the actual disposition. A precise sentence disposing of a pivotal objection can be more valuable than pages of unnecessary authorities.
Confidentiality supplies a hard boundary. Nothing in the proposal justifies feeding non-public materials into public AI services. Where no approved tool exists, the same reverse-traceability check can be performed manually—the legal safeguard should not depend on purchasing technology.
Accountability beyond automation
The strongest justification for the test, Amirkhani argues, is not administrative efficiency but the litigant's ability to understand and challenge judicial reasoning. A losing party should be able to see why an outcome-sensitive objection failed, whether the reasons were typed by a judge or refined with an approved tool.
These details were first reported by JURIST, where Amirkhani published his analysis in a personal academic capacity.