By Michael McMahon and Allison Sherwood
Forensic reviews can be the ultimate source of truth in matters involving the potential misuse of proprietary data. Courts frequently order forensic reviews—which typically involve the collection and search of electronic devices and accounts by an experienced data technology vendor—in misappropriation cases. Parties also often negotiate forensic reviews before litigation to identify and remediate misappropriation during litigation as part of discovery and during settlement talks. Indeed, forensic reviews cut to the chase: they can identify misconduct, confirm there was no misconduct, and help parties avoid or curtail costly litigation.
Parties often negotiate protocols that govern the forensic review process. These protocols require balance. An appropriate protocol will identify and set parameters for searching devices, data repositories, and accounts, while maintaining a reasonable scope to avoid unnecessary costs and over-intrusion. But that is easier said than done. Protocols are often negotiated quickly or under tight deadlines that precede injunction hearings. This can lead to disputes down the road.
The rise of AI adds a new layer of complexity to every stage of the negotiation. Because traditional protocols may miss AI-assisted exfiltration, careful attention should be paid to searching AI tools and other emerging repositories where relevant data and derivative works may reside.
Below are five practical considerations for negotiating forensic review protocols in the AI age.
Tip 1: Carefully Identify Relevant Repositories
Conventional protocols have focused on obvious repositories such as work and personal laptops, phones, tablets, external media, cloud storage services, chats, personal and work email, and collaboration tools. Parties should now also address AI platform accounts—whether enterprise-based or private—as well as services that embed AI into platforms, which may store files and retain prompts tied to the proprietary information at issue. Parties may need to move quickly, however, as data may be set for automatic deletion on a short timeline (e.g., 30 days). Using a forensic examiner with genuine experience searching AI-based repositories can be very helpful when identifying those repositories and understanding their limitations.
Early on, the parties also may not be fully familiar with all locations where exfiltrated data may reside. Accordingly, the protocol should provide flexibility and allow for the search of appropriate additional repositories in the future.
Tip 2: Tailor the Search and Reporting Methodology
Search Terms
Parties should negotiate workable search terms that are comprehensive enough to surface potential misappropriation yet tailored enough to avoid sweeping in irrelevant and burdensome material. An iterative process often works best, allowing the parties to expand or narrow search terms based on hit counts and sampling results that reveal whether, and to what extent, irrelevant documents are being captured.
To reduce costs and save time, parties may also decide to use AI-assisted review tools for hit sampling. This will require careful prompt drafting to return representative document sets that adequately inform whether existing search terms are over- or under-inclusive.
Reporting Requirements
The protocol should require delivery of forensic reports with clearly defined outputs, such as:
- Whether external devices were connected to the targeted devices.
- Whether devices were connected to cloud storage or collaboration platforms.
- Whether any mass deletions or file exports were performed, as well as audit logs identifying deletion and export activity.
- Whether the target accessed AI platforms and, if so, any prompts containing or referencing the allegedly misappropriated data along with the corresponding AI-generated outputs.
The parties should discuss the specific metadata that will be made available in the reports, as such metadata may be important to determining when and if data was exfiltrated, whether it was circulated to others, and who those recipients may be.
Tip 3: Plan Ahead to Avoid Disputes
The parties need a workable process for promptly resolving disputes before any deletion occurs. The protocol should define up front the scope of permissible objections to disclosure of specific items so the parties can address any disputes early and in good faith during meet-and-confer sessions. The protocol also should establish a predetermined window for making privilege assertions. To that end, the parties should consider how to treat potentially privileged or protected materials fed into AI tools. As a cautionary tale, in Shealy v. Seaside Investments, LLC, the court held that AI prompts and corresponding outputs generated by a litigant’s “romantic partner” are not protected under the attorney work product doctrine unless performed at the direction of counsel. No. 2684CV00799-BLS2, Dkt. No. 32, 2026 WL 2122161, at *3 (Mass. Super. Ct. June 16, 2026). Other courts following suit may deny work product protection to similar AI inputs.
Tip 4: Address Derivative Works
AI-generated derivative works—which in some cases may bear no direct resemblance to the misappropriated material—deserve unique consideration in a protocol, as the original files may have been deleted long before the forensic inspection begins. Capconvert, LLC v. Brown is a prime example of this ever-present risk. There, an employer’s internal investigation revealed that a former employee may have fed the employer’s proprietary information into an AI tool to accelerate his development of a competing company. No. 3:26-cv-02149-CRB, ECF No. 51 at 3-4, (N.D. Cal. May 26, 2026). The court noted that the employee’s rapid creation of a competing software product suggested he may have leveraged the employer’s information and specifically credited evidence that the employer’s data was stored in folders used by the employee’s AI agents. Without careful attention, traditional protocols may miss such folders.
Additionally, derivative works may result from several queries submitted over time. A functional protocol would permit the examiner to correlate AI timestamps against document access logs to build a timeline and identify potentially disparate misappropriation. The protocol also should separately account for searching prompts, outputs, timestamps, account and session identifiers, and full conversation logs, not just the AI tool’s “final” work product. This way, the protocol can target for deletion both the input (if it exists) and the resulting output or derivative work containing the misappropriated material. It may also be necessary to determine whether the AI tool is set up to learn and train itself based on inputted (and potentially proprietary) information.
Tip 5: Require Certified Deletion
A final remedial step is securing deletion of the misappropriated data. The protocol should require the examiner to use forensically sound deletion methods and to certify completion. Parties also might consider negotiating additional certifications that there are no other relevant locations that have not been searched or that there are no AI-generated derivative materials.
Key Takeaways
Protocols must be tailored to the facts of each case. Practitioners should identify the key devices and repositories at issue, tailor collection methods to those data sources, and be mindful of how AI prompts and outputs may reveal whether misappropriated material was used to create derivative works.
For a useful starting reference, see The Sedona Conference, Forensic Inspection Protocol Agreement.
Michael McMahon is a litigation partner at Cooley LLP. He has served as lead trial counsel on behalf of plaintiffs and defendants in numerous matters involving claims of trade secret misappropriation and misuse of confidential information. He regularly counsels industry-leading companies on how to protect their valuable intellectual property and works with clients to conduct detailed forensic examinations.
Allison Sherwood is a litigation associate at Cooley LLP. She has represented clients in disputes involving claims of trade secret misappropriation in federal court and arbitration, from case inception to trial,
