AI in the Courtroom

Authentication

Rule 901’s requirement that the proponent produce evidence sufficient to support a finding that an item is what it claims to be. For a photograph or recording, that has traditionally meant a witness with knowledge, and the question deepfakes raise is whether that is still enough.

Machine-Generated Evidence

Output produced by a process or system rather than a human, such as a software analysis of cell-site data or an AI-generated summary of a video. When no expert testifies to how it was produced, the court has no one to cross-examine about its reliability, which is the gap proposed Rule 707 was written to close.

Expert Reliability

Rule 702’s requirement that expert testimony rest on sufficient facts, reliable methods, and a reliable application of those methods to the case. Proposed Rule 707 would apply the same test to a machine’s output offered in place of an expert.

Deepfake Detection Limits

The technical reality that detection tools produce probabilities rather than proof, lag behind generation tools, and fail on compressed or re-encoded files. A lawyer who promises a jury that software can tell real from fake is promising more than the science delivers.

Notice and Removal

The TAKE IT DOWN Act’s requirement that a covered platform maintain a process for a victim to request removal of a nonconsensual intimate image, real or synthetic, and remove it and known copies within 48 hours. The FTC enforces the duty as an unfair or deceptive practice.

Civil Remedies

Private causes of action for people depicted in sexual deepfakes, which the federal DEFIANCE Act would create and which Florida’s § 836.13 already provides alongside criminal penalties. Takedown stops the spread; a civil remedy is how the victim gets paid.

Judicial Use of AI

Court policies governing when judges and staff may use generative AI, typically permitting summaries, drafting help, and editing while requiring that the judge personally own every ruling, verify every citation, and keep confidential information out of public tools. New York’s interim policy and its January 2026 guidance are the most developed examples, and Part 4 grades a draft order against them.

Part 1: Read the Rules

Read Rule 901 and Rule 702, the Rule 707 memo in the May 2026 agenda book, and the CRS explanation of the TAKE IT DOWN Act. Then read the New York interim policy. You should be able to state what a proponent must show to authenticate a recording, what proposed Rule 707 would add, what a platform must do within 48 hours, and what a judge may and may not delegate. Because the rule’s status changes with each committee meeting, use the Rule Tracker to confirm where Rule 707 stands today and check its answer against the uscourts.gov page it points you to.

Part 2: Authenticate Your Own Deepfake

Using any AI tool, create a benign altered file of yourself: a photo with the background or an object changed, or a short audio clip of a synthetic version of your voice reading a paragraph you wrote. Do not create anything involving another person. Imagine the file is offered as evidence in a civil case, and use the Foundation and Objection Builder to write the questions a proponent would ask a witness to lay the foundation under Rule 901, the objections an opponent would raise, and what each side would need from a technical witness under Rule 702. Include a paragraph on whether proposed Rule 707, if adopted, would change the analysis.

Part 3: Draft the Takedown Notice

Take the hypothetical below and use the Takedown Notice Drafter to write a removal request to the platform that satisfies the TAKE IT DOWN Act’s requirements as the FTC describes them, and a short cover note to the client explaining what the platform must do, by when, and what happens if it does not. Check every statutory element the AI includes against the CRS summary and the statute.

Hypothetical: Your firm’s client, a 24-year-old Daytona State graduate, learns that a former acquaintance has posted AI-generated sexual images bearing her face to a social media platform and a message board. She has screenshots with URLs and dates. She wants the images down today and wants to know what else the law offers her.

Part 4: Grade a Judge’s AI Draft

Ask any AI tool to draft a two-page order granting a routine motion to extend a discovery deadline in a fictional Florida circuit court case, with a short statement of reasons and a citation to the governing rule. Then use the Chambers Reviewer to grade that draft against the New York interim policy and the January 2026 guidance: what a judge could use as drafted, what must be personally verified, what should never have gone into a public tool, and what the judge would need to disclose. Verify the rule citation yourself.

Part 5: Post Your Deliverables

Create a single new discussion thread containing your altered file with a one-line description of what you changed, your foundation questions and objections with the Rule 707 paragraph, your takedown notice and client note, and your graded order with the reviewer’s findings and your own corrections. Close with a paragraph on which of the three fronts you think the law has handled best so far.

Part 6: Share Chat Link

Include one AI chat link with a 1–2 sentence explanation of what the conversation shows and why you chose to share it.

Foundation and Objection Builder

Act as a trial evidence professor. I will describe a file, a photograph or audio recording of myself that I altered with an AI tool, and the civil case in which it is offered. First, write the questions the proponent’s lawyer would ask a witness to lay a foundation under Federal Rule of Evidence 901, using the illustrations in 901(b). Second, write the objections the opposing lawyer would raise and the questions on voir dire that would probe whether the file was altered. Third, explain what each side would need from a technical witness under Rule 702 and whether any detection tool could resolve the dispute. Finally, explain whether proposed Federal Rule of Evidence 707, as published for comment, would change the analysis, and say plainly that the rule is not in force. Cite the rule text so I can check it.

Both sides of the foundation. Writing the direct examination and the voir dire for the same file shows you what authentication actually turns on, and the instruction to note that Rule 707 is only proposed keeps a pending rule from being cited as law.

Takedown Notice Drafter

Act as a paralegal at a firm that represents victims of nonconsensual intimate imagery. I will describe a client’s situation and the platforms involved. Draft a removal request to each platform that satisfies the TAKE IT DOWN Act’s notice requirements as the FTC has described them: identification of the depicted person, a good-faith statement that the image is nonconsensual, the URLs, contact information, and a signature. Then draft a short note to the client explaining the platform’s 48-hour obligation, the FTC’s role if the platform fails to act, and the civil and criminal remedies that may also be available under Florida law and pending federal legislation. Cite the statute or guidance for every requirement so I can verify it, and tell me if any element is one you are unsure the law actually requires.

A document built from statutory elements. Requiring a citation for each element makes the notice checkable against the CRS summary, and the client note forces the harder question of what the law offers beyond removal.

Chambers Reviewer

Act as an ethics advisor to a state court judge. I will paste a draft court order that a judge generated with an AI tool, along with the New York State Unified Court System’s interim AI policy. Grade the draft against the policy: identify what the judge could use as written, what the judge must personally verify before signing (including every citation), what information should never have been entered into a public AI tool, whether any disclosure to the parties is required, and whether anything in the draft reads as the machine’s judgment rather than the judge’s. Cite the specific policy provisions you rely on. Do not rewrite the order; evaluate it.

Applying a court’s own rules to a court’s own draft. Grounding the review in the pasted policy rather than the model’s sense of judicial ethics produces findings you can trace to a paragraph, which is how a real ethics advisor would answer.

Rule Tracker

Act as a federal rules librarian. Explain where proposed Federal Rule of Evidence 707 stands in the federal rulemaking process as of the date I give you: what the rule would do, what the Advisory Committee on Evidence Rules decided at its most recent meeting, what steps remain before any rule takes effect, and the earliest date it could apply. Where your information may be out of date, say so and tell me which page on uscourts.gov to check for the current status. Do not describe the rule as in force.

A status question with a stale-knowledge warning. Rulemaking moves in dated steps, so the prompt asks for the sequence and the source to check rather than a confident answer, which is the right shape for any question about a pending rule.