--- title: "Incoming evidence — deepfakes & synthetic documents" url: "https://ailawframework.com/obligations/incoming-evidence-deepfakes-and-synthetic-documents" type: "obligation" site: "AI Law Framework" id: "A44" task_count: 5 last_reviewed: "2026-09-28" --- # Incoming evidence — deepfakes & synthetic documents > Preserve original material and establish its provenance before relying on it. Document handling and metadata, and obtain suitable forensic assistance where authenticity is disputed or material to the matter. Canonical page: https://ailawframework.com/obligations/incoming-evidence-deepfakes-and-synthetic-documents Last reviewed: 2026-09-28 Site index for agents: https://ailawframework.com/llms.txt This policy mostly governs the AI output the firm produces. AI also threatens evidence coming in: fabricated documents, deepfake audio/video/images and synthetic 'discovered' material. Authenticity cannot be assumed. ## Key facts - Tasks: 5 - Green: 0, Yellow: 3, Red: 2 - Levels: Green is AI drafts and the lawyer checks; Yellow is lawyer and AI draft together; Red is the lawyer writes and AI may only check, clarify or collect data. ## Principles - **Define the matter**: Confirm the client’s instructions, jurisdiction, intended output and the information available before dividing the work. - **Confirm authenticity**: Preserve original files and relevant metadata. Document provenance and handling, and seek forensic review where authenticity needs to be resolved. - **Confirm the source material**: Check the work against the original records and identify omissions, contradictions and unresolved information. ## Checklist - Preserve original files and relevant metadata. Document provenance and handling, and seek forensic review where authenticity needs to be resolved. - Resolve exceptions and record the responsible lawyer’s approval of the exact output and intended use. ## Restrictions - Confirm firm, client and applicable court or professional restrictions before using any AI feature. ## Jurisdiction note Apply this guidance to the actual matter and jurisdiction. Confirm current requirements before relying on an example. ## Basis Practice guidance supports planning; the responsible lawyer confirms the scope, evidence and review requirements for the matter. This policy mostly governs the AI output the firm produces. AI also threatens evidence coming in: fabricated documents, deepfake audio/video/images and synthetic 'discovered' material. Authenticity cannot be assumed. ## Limits on AI assistance Red lines — never in this area ## Limits on AI assistance Never rely on an AI 'deepfake detector' or AI-asserted provenance as the sole basis for accepting or tendering evidence. Raise authenticity with the court/opponent where material evidence is suspected synthetic. Preserve originals and metadata; document the authentication steps taken. - Never rely on an AI 'deepfake detector' or AI-asserted provenance as the sole basis for accepting or tendering evidence. - Raise authenticity with the court/opponent where material evidence is suspected synthetic. - Preserve originals and metadata; document the authentication steps taken. ## Review checklist The lawyer's review checklist ## Review checklist Provenance of pivotal evidence confirmed Metadata preserved & reviewed Forensic check where authenticity is in issue Chain of custody documented - Provenance of pivotal evidence confirmed - Metadata preserved & reviewed - Forensic check where authenticity is in issue - Chain of custody documented ## Tasks (5) | Task | Level | How AI is used | What the lawyer must review | | --- | --- | --- | --- | | [Build an authenticity checklist for a key digital exhibit](https://ailawframework.com/obligations/incoming-evidence-deepfakes-and-synthetic-documents/build-an-authenticity-checklist-for-a-key-digital-exhibit.md) | Red | From the file, list source, capture date, metadata fields, hash if supplied and gaps against the authenticity questions the lawyer has named. | Authenticate through source and chain of custody, using a forensic examiner where needed Do not rely on a tool's 'this looks real' score | | Flag candidate synthetic or altered items in a production set | Yellow | From a permitted production, list items with missing metadata, inconsistent fonts or reuse patterns, and a sample for human review. | Spot-check pivotal documents Treat a generated 'synthetic' flag as a question, not a finding | | Compare claimed source metadata with the file actually produced | Yellow | Extract stated author, date and device fields and compare them with the produced file's properties, listing mismatches. | Treat a match as not proving authenticity: metadata can be edited Enquire further on any mismatch | | Draft an authenticity challenge or expert-instruction letter | Red | The tool tabulates the disputed exhibits against the authenticity checklist, the metadata mismatches and the chain-of-custody gaps, and indexes the native… | The lawyer writes both documents. Before alleging that an exhibit is fabricated or synthetic they satisfy themselves there are reasonable grounds on material… | | Build a chain-of-custody log from the capture papers | Yellow | From the capture notes, list who held the device or file, when, and what hashes or export logs exist, marking gaps. | Do not let the log invent a custodian Treat a broken chain as going to weight | ## Jurisdiction notes ### New Zealand #### Court guidance The Courts of New Zealand lawyer guidelines warn that tools such as DALL·E, Midjourney and Stable Diffusion can create images, audio or video that look authentic but are fake. #### Risks Deepfakes can be defamatory or spread misinformation. Authenticity concerns arise when multimedia is used in evidence. #### Your duty You remain responsible for the accuracy of material before the court. Do not enter non-public evidence into a public chatbot to test it. - Do not enter private, confidential, suppressed or privileged material into a public chatbot. - Do not tender generated multimedia as authentic evidence. Sources: - [Courts of New Zealand: GenAI guidelines — Lawyers (7 December 2023)](https://www.courtsofnz.govt.nz/assets/6-Going-to-Court/practice-directions/practice-guidelines/all-benches/20231207-GenAI-Guidelines-Lawyers.pdf) (checked 2026-09-23) - [Defamation Act 1992](https://www.legislation.govt.nz/act/public/1992/105/en/latest/) (checked 2026-09-23) ### United States Local name: Incoming evidence — deepfakes and synthetic documents #### Proposed FRE 707 Not recommended for action on 7 May 2026. The committee revised the draft on AI evidence offered without an expert and will study it further. Not in force. #### Deepfakes The committee is still studying deepfake evidence. No new rule is in force. #### Current law Authenticity and expert reliability stay under existing Articles 7 and 9 and case law. Formal Opinion 512 applies to tools used to screen media. - Do not tell a court that proposed Rule 707 is in force. It was not advanced on 7 May 2026. Sources: - [Advisory Committee on Evidence Rules, 7 May 2026 meeting report (proposed Rule 707)](https://www.uscourts.gov/sites/default/files/document/advisory_committee_on_evidence_rules_may_2026.pdf) (checked 2026-09-23) - [ABA Formal Opinion 512, Generative Artificial Intelligence Tools (29 July 2024)](https://www.americanbar.org/content/dam/aba/administrative/professional_responsibility/ethics-opinions/aba-formal-opinion-512.pdf) (checked 2026-09-23) *Practice guidance for planning and review. Confirm current primary sources and the requirements of the actual matter. This is general information, not legal advice.*