# AI-assisted FCA evidence review.

> Canonical URL: https://getwolf.ai/guides/ai-assisted-fca-evidence-review
> Published: 2026-08-23
> Last reviewed: 2026-08-24

AI-assisted FCA evidence review uses software to surface potentially relevant passages and organize them for a lawyer, while keeping every proposed finding tied to the source material for validation.

Written and maintained by the WOLF AI product team and checked against the public sources cited on this page. It has not been reviewed by outside counsel, and it is not legal advice.

WOLF AI is a software company, not a law firm. This page is educational and is not legal advice.

## What do the numbers say?

**Model Rule 1.1 — Duty of technology competence.** Comment 8 to ABA Model Rule 1.1 states that maintaining competence requires a lawyer to keep abreast of "the benefits and risks associated with relevant technology." Adopting an AI-assisted workflow and verifying what it produces are treated as two halves of the same professional obligation, not as separate choices.

Source: [ABA Model Rule 1.1: Competence (including technology)](https://www.americanbar.org/groups/professional_responsibility/publications/model_rules_of_professional_conduct/rule_1_1_competence/)

**NIST AI RMF — Federal framework for AI risk.** The NIST AI Risk Management Framework organizes trustworthy AI around functions it calls Govern, Map, Measure, and Manage, and treats traceability and human oversight as design requirements rather than optional additions. Source-linked output is the practical form that traceability takes in a litigation workflow.

Source: [NIST AI Risk Management Framework](https://www.nist.gov/itl/ai-risk-management-framework)

**Rule 9(b) — Particularity standard for pleading fraud.** Federal Rule of Civil Procedure 9(b) requires that a party "must state with particularity the circumstances constituting fraud or mistake." In practice that is what turns an FCA matter into a document-organization problem: the allegation has to name specific claims, dates, and actors, each traceable to a record.

Source: [Federal Rule of Civil Procedure 9(b): pleading fraud with particularity](https://www.law.cornell.edu/rules/frcp/rule_9)

## Who is this workflow for?

Associates and staff attorneys carrying first-pass review on evidence-heavy FCA matters, and the partners who have to trust that pass without re-reading all of it themselves.

**When it is the wrong tool.** It is not appropriate as an unreviewed filter. A workflow in which flagged material goes forward and unflagged material is never looked at again puts a recall problem the software cannot see in the middle of the case.

## What documents and inputs do you need?

- PDFs, including scanned records processed with OCR.
- A stated review scope: custodians, date range, and the conduct in question.
- Any coding decisions or issue tags the team already works from, so candidate material lands in the team’s own vocabulary.

## How does the workflow actually run?

1. Ingest PDFs, including scanned records processed with OCR.
2. Extract candidate entities, dates, billing references, and passages.
3. Present flagged material with its document, page, and passage location.
4. Let the legal team edit, reject, annotate, and approve the resulting work product.

## What does this look like in practice?

**Illustrative scenario.** Illustrative only. A review set contains 1,800 pages of scanned clinical notes plus a billing export converted to PDF.

1. OCR runs on the scans; pages where confidence is low are marked rather than silently accepted.
2. Candidate passages are grouped by the entity and date they mention, not by the file they happen to live in.
3. A reviewer works the group, not the folder: every claim line referencing one provider on one date arrives together with the note that is supposed to support it.
4. Rejections are recorded, so the next pass does not re-surface material a human already ruled out.

**What this does not show.** A low-confidence OCR page that nobody opens is a gap in the review, not a clean result. Treat the marked pages as a worklist.

## What do you get out, and who reviews it?

- A reviewed set of candidate passages, each with its source location.
- A record of what was accepted, rejected, and annotated, and by whom.
- Counsel owns the recall question — whether the material that was never flagged still needs eyes on it.

## Where does this approach break down?

- A flagged passage is not a legal conclusion or proof of a false claim.
- OCR and extraction can be wrong and must be checked against the original document.
- Native support for formats beyond PDF should be confirmed during a demo.
- The software reports what it surfaced. It cannot report what it missed.

## Common questions

**Does the software decide what is relevant?**

It proposes candidates against the stated scope. Relevance, privilege, and significance are decided by the legal team.

**What happens to material that is never flagged?**

It stays in the workspace and remains searchable. Deciding how much unflagged material still needs human review is a scoping decision for counsel, not a software setting.

**Is client data used to train models?**

Current data-handling practice, retention, and access limits are described on the security page and walked through directly during a demo.

## Where do these facts come from?

- [U.S. Department of Justice: The False Claims Act](https://www.justice.gov/civil/false-claims-act)
- [NIST AI Risk Management Framework](https://www.nist.gov/itl/ai-risk-management-framework)
- [ABA Model Rule 1.1: Competence (including technology)](https://www.americanbar.org/groups/professional_responsibility/publications/model_rules_of_professional_conduct/rule_1_1_competence/)

## Continue

- [See evidence-review outputs](https://getwolf.ai/products#features)
- [How WOLF AI pricing works](https://getwolf.ai/pricing)
- [Book a demo](https://getwolf.ai/book-demo)
- [Security and data handling](https://getwolf.ai/security)

## Related field guides

- [A source-linked qui tam chronology builder](https://getwolf.ai/guides/qui-tam-chronology-builder)
- [Methodology for source-linked outputs](https://getwolf.ai/guides/methodology-source-linked-outputs)