//Counsel whose evidence AI has touched
AI Evidence Consulting Expert for Attorneys
Matt Aubin works behind counsel as the consulting expert on AI assisted and AI generated evidence. He validates it or attacks it, writes the analysis, and briefs the attorney. When a firm needs its own AI use checked and signed for its insurer, the same checks come back as a signed AI evidence audit.
By Matt Aubin, Founder, Southern Recon Agency and E3 Legacy Intel. CDFE, FBCI. Investigating since 2009. Updated .
Status as of September 2026
Proposed Rule 707 and draft Rule 901(c) are not in effect. Both go before the Advisory Committee on Evidence Rules mini conference on October 15, 2026. The reliability test they point to, Rule 702, already applies today. Full status, with primary sources, on the two rule page.
When AI touched the evidence, somebody has to be able to say whether the machine was reliable. That is the consulting expert's job, and it is done for the attorney, not for the other side.
What does a consulting expert on AI evidence do that a disclosed expert does not?
A consulting expert works behind the attorney and is not disclosed to the other side. Matt Aubin takes that seat. He examines the AI assisted or AI generated evidence and tells counsel plainly whether it holds up. He writes the analysis and briefs the attorney on what to demand in discovery, what to ask, and where the method breaks. Counsel keeps control of what is filed and of who, if anyone, is put forward to the court. The product is a written analysis for the attorney, not a report the other side gets to pick apart.
What does an AI evidence expert do for counsel before trial?
Reads the evidence early, while there is still time to do something about it. An AI evidence expert tells counsel what the AI actually produced, what record exists behind it, and whether that record survives a Rule 702 reliability challenge. The work is the written analysis, the discovery and question list, and the briefing, all of it behind the attorney.
The point of doing it before trial is that almost everything that decides an AI evidence fight is a discovery fact: the inputs, the prompts, the system and version, the human verification, the chain of custody. Those are asked for in discovery or they are not asked for at all. Matt Aubin does not present as a witness and is not offered as one. Counsel keeps control of what is filed and of who, if anybody, goes in front of the court.
Does consulting expert privilege protect the analysis?
Usually, and that is the point of the seat. Federal Rule of Civil Procedure 26(b)(4)(D) ordinarily shields the facts known and opinions held by an expert retained in anticipation of litigation who is not expected to be called as a witness at trial. The protection is not absolute, and state rules differ, so how the retention is papered matters. This is general information, not legal advice.
Two things follow from that in practice. Retain through counsel rather than directly, so the work sits inside the attorney's file from the first day. And decide early, in writing, whether this expert is ever going to be disclosed, because the answer changes what is protected. Read the rule itself: Federal Rule of Civil Procedure 26.
How do you validate an AI assisted investigation before it reaches court?
Four checks, mapped to the Evidence-Grade AI Standard. Every finding traces to a source that exists and says what the report says it says. A supervision log shows what was asked of the AI, what came back, and who verified it. The records behind each finding are checked against the databases they came from. And the method's limits and error rate are stated rather than assumed. The system and version, the steps, and the chain of custody are documented so another examiner could repeat the work. That record is what a Rule 702 reliability challenge asks for.
The full framework is the Evidence-Grade AI Standard. When the question is your own firm's workflow rather than a single piece of evidence, the AI Evidence Audit runs the same checks across the whole process and delivers a signed document the firm can hand its insurer.
How do you attack the other side's AI generated or deepfake evidence?
The same checks, run in reverse. Start with the record. If the other side cannot produce the inputs, the prompts, the system and version, and the human verification behind an AI output, that output is an assertion, not a finding. For media offered as real, the questions are provenance, metadata, and the artifacts a generative model leaves behind. For a report written with AI, every citation and every figure is checked against the original. Counsel receives a written analysis: what to demand in discovery, which questions to put to the sponsoring witness, and where the method fails.
Deepfake examinations themselves, the forensic work on the video, audio or image, run through E3 Legacy Intel at e3intel.io/deepfake-verification. This site covers the consulting role behind counsel; the deepfake consulting expert page explains how the two fit together.
Send the evidence and the deadline.
The exhibit, the report or the expert disclosure. I read it and come back with what it is and what it will take.
What do proposed Rule 707 and draft Rule 901(c) change, and what is their status today?
Neither is in effect as of September 2026. Rule 707 would hold machine generated evidence to the same reliability standard Rule 702 applies to expert opinion. Draft Rule 901(c) would set the test for evidence a party claims AI fabricated. Both go before the Advisory Committee's mini conference on October 15, 2026.
The December 1, 2027 effective date still circulating online is outdated. Law firm posts and university pages are still publishing that Rule 707 was approved and takes effect on that date. It was approved for publication for public comment, which is not adoption. The committee record then shows the rule revised and deferred for further study on May 7, 2026, and the Standing Committee reporting in June 2026 that it does not recommend action at this time.
Rule 707 went through public comment, which closed February 16, 2026, and drew more than seventy written comments and a public hearing. Draft Rule 901(c) was never published for comment at all. The committee drafted the text and decided in May 2026 that an amendment was not needed yet, keeping the draft ready.
As drafted, Rule 901(c) works in two steps. The opponent must first offer enough for a reasonable person to infer fabrication. Only then must the proponent prove authenticity to a higher standard than Rule 901 normally asks.
The test both rules point to, Rule 702, already governs expert evidence today, which is why the reliability record is worth building now rather than after a rule lands. This is general information, not legal advice.
The dated status of both rules, with links to the committee's own materials, is kept current on Rule 707 and Rule 901(c): AI evidence status. Primary sources: the rules committee agenda books on uscourts.gov and the text of Rule 702 and Rule 901.
What does Florida Bar Ethics Opinion 24-1 require of the lawyer using AI?
A lawyer may use generative AI, subject to four conditions. Protect client confidentiality. Supervise the AI's output the way the rules require supervision of a nonlawyer assistant. Keep fees and costs reasonable, so the client is not billed for time the AI saved. Follow the advertising rules if AI is used in marketing. In practice the lawyer must verify what the AI produced before relying on it, and cannot hand that responsibility to the tool. This is general information, not legal advice.
That verification is work a consulting expert can do and document for the file, which is the analysis Matt Aubin writes. Read the opinion itself: Florida Bar Ethics Opinion 24-1.
What happens when an expert's report was written with AI (the Kohls v. Ellison problem)?
The court struck an expert declaration about deepfakes because it cited sources that did not exist. The expert had drafted with a generative AI tool and had not caught the fabricated citations. The lesson is that an AI drafted expert report is only as reliable as the human check behind it. Before your own expert's report goes out, every citation and every figure should be verified against the original by somebody other than the author. When the other side's expert used AI, the same check is how you find the holes.
Both are consulting work Matt Aubin does for counsel, documented so the attorney can rely on it. The docket is on the government's own site: Kohls v. Ellison, No. 24-cv-03754 (D. Minn.).
How are fees set?
Every engagement is priced on the value of the work, scoped in writing after one conversation about the evidence, the volume, and the deadline. The retainer is paid in full before work begins, and the scope states exactly what counsel receives: the written analysis, the discovery and question list, and the briefing. There is no hourly meter and no surprise invoice. If the matter grows, a new scope is agreed before the added work starts.
//Why this seat
Why counsel puts Matt Aubin behind the attorney on AI evidence
He holds the three things the seat needs at once. Investigating since 2009. Founded his first firm in 2010 and built it into Southern Recon Agency in 2014. The Florida Board Certified Investigator designation, held by fewer than fifty people in Florida, and the Certified Digital Forensics Examiner credential, so the evidence is handled to a forensic standard. And the code: he built the proprietary AI investigation platform his own firms run on live cases.
On AI evidence he works as the consulting expert behind counsel, which keeps the analysis privileged and the strategy in the attorney's hands. He also teaches the subject: he trained a delegation of Brazilian prosecutors at World Orlando, the State Department sponsored conference in Orlando, and he delivered "A-I for the Modern Private Investigator" at Intellenet's 2026 conference on April 15, 2026.
When the question is a firm's own AI use rather than one piece of evidence, the same checks are delivered as the signed AI evidence audit, a signed document the firm can hand its insurer.
If the question underneath the matter is ownership rather than reliability, the primary sources are laid out on who owns work that AI helped make. If the matter is a live wire fraud, a business email compromise or a deepfake used to move money, start at AI cybercrime consulting instead, and the rule making behind AI evidence is tracked on the Rule 707 page.
Send the evidence and the deadline; you get a written analysis for counsel.
Tell me what was produced, how AI touched it, and when the motion is due. I will scope it in writing the same conversation.