Product

Solutions

Security

Company

Resources

Request a Demo

Product

Solutions

Security

Company

Resources

Request a Demo

Article

2026-09-25

Proposed Rule 707: AI-generated evidence and what courts may require

The Advisory Committee on Evidence Rules did not advance the published version of proposed Rule 707 and is studying a revised draft aimed at AI-generated evidence offered without an expert.

At the pretrial conference in a construction defect case, counsel discloses an AI-generated reconstruction showing water moving behind the building’s exterior wall. The proposed witness is the project manager who uploaded the inspection records. Opposing counsel asks for the input files and an explanation of how the model determined the water’s path.

Proposed Federal Rule of Evidence 707 would apply Rule 702-style reliability requirements to AI-generated evidence offered without an expert when its substance would otherwise require expert testimony. It is not law. The committee is studying a revised draft after declining to advance the published version. At Turbo Law, we link case facts to their source pages; the underlying records remain the evidence.

What is proposed Federal Rule of Evidence 707?

The published proposal sought to close a gap: machine-generated conclusions could reach the courtroom without the reliability scrutiny that would apply if an expert delivered the same opinion. The Standing Committee approved publication in June 2025, and the draft appeared for public comment in August 2025.

Its trigger mattered. Under that draft, machine-generated evidence offered without an expert would face Federal Rule of Evidence 702(a) through (d) requirements if testimony expressing the same substance would be subject to Rule 702. The Advisory Committee on Evidence Rules’ May 2026 report explains that concern.

Think about the project manager. Knowing which inspection photographs went into the program does not necessarily qualify that witness to explain a reconstruction of moisture migration through concealed wall assemblies.

The missing explanation matters.

A witness might competently identify the photographs while having no basis to defend the program’s assumptions about conditions inside the wall.

Where does proposed Rule 707 stand now?

As of September 25, 2026, proposed Rule 707 remains under study. The committee did not advance the published draft or recommend republication at its May 7, 2026 meeting.

Public hearings were held on January 15 and January 29, 2026; comments closed February 16, 2026. Responses were mixed. Criticism included the breadth of “machine-generated evidence”, a phrase that raised questions about how far the proposal would reach beyond the AI outputs prompting concern.

The leading redraft focuses on artificial intelligence and carries the title “Evidence Produced by Artificial Intelligence and Presented at Trial Without an Expert”. It adds notice and ordinarily calls for an expert foundation, while leaving room for other reliability proof in exceptional circumstances.

That draft is still a proposal.

The October 2026 agenda book, posted September 23, 2026, includes the revised proposal and a separate deepfakes proposal for further study, with an expert mini-conference planned. The committee plans to take it up again this fall. No outcome should be assumed.

What is proposed Federal Rule of Evidence 707?

The missing explanation matters.

A witness might competently identify the photographs while having no basis to defend the program’s assumptions about conditions inside the wall.

Where does proposed Rule 707 stand now?

As of September 25, 2026, proposed Rule 707 remains under study. The committee did not advance the published draft or recommend republication at its May 7, 2026 meeting.

That draft is still a proposal.

How would a proposed Rule 707 foundation work in practice?

The redraft would require the proponent to establish helpfulness to the factfinder, sufficient supporting facts or data, reliable principles and methods, and reliable application to the case. It would ordinarily require an expert to explain how the system reliably produced the evidence, even though the output itself is being offered without an expert witness.

Other proof could suffice in exceptional circumstances. The draft also requires reasonable notice that gives the opposing party a fair opportunity to meet the evidence.

For the water reconstruction, we would organize preparation around the following questions. These are practical requests to consider, rather than a disclosure checklist already imposed by proposed Rule 707:

What will the exhibit prove? Identify the disputed proposition. A reconstruction offered to establish where water entered the wall needs a foundation directed to that inference.
What went into the model? Preserve the actual input files. Compare the inspection photographs with the repair records and identify any assumptions supplied where observations were missing.
Why does the method fit this task? Seek documentation or validation addressing the use at issue. A general description of the software may say little about reconstructing moisture movement in this building.
How was the method applied here? Keep the settings and model version with the output. Document edits, including any change to the assumed condition of concealed flashing.
Who can explain it, and when? Identify the proposed foundation witness and arrange disclosure early enough for a meaningful challenge. If relying on other proof, identify the exceptional circumstances and the reliability showing the court would be asked to accept.

Start with the input folder.

An opponent reviewing the reconstruction should be able to distinguish an observed condition from an assumption entered by the operator, particularly where the animation makes both look equally certain.

How do courts handle AI-generated evidence today, without proposed Rule 707?

Existing evidence rules still govern, with the applicable foundation depending on what the exhibit purports to establish and how it is offered. A pending proposal supplies no present shortcut to admission or exclusion.

Under Federal Rule of Evidence 702, amended effective December 1, 2023, the proponent must demonstrate that it is more likely than not that expert testimony satisfies the rule’s requirements. An expert’s reliance on AI does not remove the court’s responsibility to examine the method and its application. Counsel should be prepared to explain that reliance before the witness takes the stand.

Authentication remains a separate issue. Federal Rule of Evidence 901(a) requires evidence sufficient to support a finding that the item is what its proponent claims. Subsection (b)(9) addresses evidence describing a process or system and showing that it produces an accurate result. Identifying a file as the program’s output leaves open whether its substantive conclusions have an adequate foundation.

Federal Rule of Evidence 1006 permits summaries of voluminous admissible materials that cannot conveniently be examined in court. The underlying materials need not themselves have been introduced, but originals or duplicates must be available to the other parties for examination or copying at a reasonable time and place. A qualifying summary is itself evidence.

Calling an AI reconstruction a summary does not establish that it merely reports the contents of records.

We would raise the intended use before the exhibit deadline and seek a schedule for foundation disclosures and objections. State-court litigation requires a separate check of the governing jurisdiction’s rules.

Is using AI to organize a case file the same as offering AI evidence?

No. Using AI to locate a repair email or assemble a working chronology does not, by itself, put the AI’s output before the factfinder as proof.

Consider a chronology entry saying the owner reported staining before the contractor opened the wall. Counsel can follow the entry to the email, read the surrounding exchange and decide whether that email supports the point. The chronology may remain an internal working document throughout trial.

The distinction can disappear quickly.

Suppose counsel instead offers a generated narrative asserting that the staining proves a concealed installation defect. The software has now supplied an inference that needs its own scrutiny. A link to the staining email does not establish the cause.

An AI-assisted chart offered under Rule 1006 also needs the applicable foundation, regardless of how useful it was during preparation. We would review every entry against the underlying materials and identify any wording that adds an inference beyond what those materials show.

How does source-linked case work keep the record front and center?

It lets the team return to the document behind a factual statement while deciding what can actually be proved. That is where our work sits.

Turbo Law Review reads the whole matter file, including pleadings and productions, correspondence, deposition transcripts, expert reports and work the team has already produced. We build a connected picture of the case with each fact linked to its source page and line. New records update that picture; chronologies can flag gaps where records appear to be missing.

For the construction file, a useful assignment might compare the project manager’s account of the first leak with the inspection report. The lawyer can open the cited passage and check whether the report describes an observation or repeats something the owner said.

That difference may change the examination.

Our Assistant returns line-cited work product for attorney review and says when the file does not support an answer. Its exportable assignment log records the request, the material read and the resulting work product. It prepares work; it does not serve, file or send.

Review and drafting start from the same case picture, as do research and reporting. The firm’s document management system remains the system of record.

We do not create evidence or give admissibility opinions. Lawyers decide what to offer after checking the source and the foundation required for that particular use.

FAQ

Is proposed Rule 707 in effect?

It is not in effect as of September 25, 2026. The published draft did not advance at the May 2026 meeting, and the revised text remains under study; a motion should distinguish that proposal from the evidence rules currently governing the case.

Does proposed Rule 707 cover deepfakes?

The committee is studying deepfakes separately. Proposed Rule 707 addresses reliability concerns where evidence is acknowledged to be AI-produced and would require expert treatment if expressed by a witness. A fabricated recording presented as genuine raises an authenticity problem, although a particular exhibit could raise both issues. The companion deepfakes proposal also remains under study.

Do I need an expert to admit AI-generated evidence?

There is no universal answer based solely on the AI label. Current law requires attention to the exhibit’s substance and intended use, while the revised proposed Rule 707 ordinarily contemplates an expert foundation for the outputs within its scope. Identify who can explain the process before committing to the exhibit.

When could proposed Rule 707 take effect?

No effective date can responsibly be assigned. The U.S. Courts’ rulemaking overview describes an amendment as generally taking about three years, but that description does not establish a timetable for this proposal. For a pending case, put the proposed exhibit and its foundation on the next pretrial agenda under the rules already in force.