Around October 7, 2026 the Federal Trade Commission began issuing investigational subpoenas (Civil Investigative Demands and merger Second Requests) that impose significantly stricter requirements on the use of technology-assisted review (TAR) or AI-assisted review. The new instructions demand expansive upfront disclosures, minimum precision and recall thresholds, complete collections before training, and disclosure of prompts and rationales. Legal and e-discovery teams responding to FTC matters should review the new conditions immediately and decide whether to negotiate, modify protocols or revert to traditional methods.
The FTC has issued subpoenas containing far more rigorous rules for using AI or TAR to identify responsive documents. In-house counsel, outside counsel and e-discovery managers handling FTC investigations or Second Requests need to treat these instructions as a material change in compliance burden and timing.
This is not legal advice. Confirm requirements with counsel on any specific matter.
- What changed? FTC investigational subpoenas now include expanded, prescriptive requirements for TAR and AI-assisted review, including minimum precision/recall metrics, pre-training disclosures and prompt disclosure.
- When? The stricter instructions appeared in recent Civil Investigative Demands and Second Requests; analysis published October 7, 2026.
- Who is affected? Companies and counsel responding to FTC investigations or merger Second Requests that plan to use TAR or AI for document review.
- What to do this week? Obtain the full instructions from any open FTC matter, assess whether current protocols meet the new thresholds, and decide on negotiation or alternative review methods.
What new requirements has the FTC introduced?
According to analysis of recent FTC subpoenas, the Commission is now requiring expansive upfront disclosures at least one week before any model training or review begins. These include a detailed description of the technology and workflow, identification of all subject-matter experts (who must be full-time employees of the responding firm), reports on document types and exclusions, and—if AI is used—the tool, version, provider, underlying model and AI-versus-human agreement rates on a statistically valid sample. The FTC also requires parties to meet minimum performance thresholds: recall of at least 75 percent (margin of error ±3 percent at 95 percent confidence) and, for the first time, precision of at least 75 percent, with a disclosed cutoff score no lower than 0.50. Collections from all custodians must be complete before training can begin. Parties must also disclose prompts, relevance scores and rationales if available, provide control and training sets, run an elusion validation sample for FTC review at least two weeks before compliance, and supply a calendar of anticipated productions within a week of training start. Pre-culling via analytics without FTC agreement is prohibited, and manual review or search terms may not be used to remove documents identified as responsive except for privilege or sensitive personal information.
How do these rules differ from prior FTC practice?
Previous FTC instructions typically required a written description of any technology used to identify or eliminate responsive documents. The new instructions expand that requirement substantially, add numerical performance floors (including precision), formalize validation sampling, insert the FTC into production scheduling, and require disclosure of prompts and rationales. The complete-collection-before-training rule is expected to delay the start of TAR or AI and appears aimed at discouraging older TAR workflows that need a large upfront corpus. The precision requirement is a departure from earlier agency focus on recall alone.
Why are these changes appearing now?
The FTC’s instructions explicitly define AI-assisted review for the first time, reflecting its increasing use by parties. The tighter controls give the Commission greater visibility into, and influence over, the review process. They also raise the practical cost and time required to use advanced review technology, which may lead some parties to revert to search-term and manual review for smaller populations or to negotiate modifications. The changes arrive while AI use in e-discovery is expanding and while courts continue to address related issues such as the discoverability of AI prompts.
What should legal and e-discovery teams do this week?
If you have an open FTC Civil Investigative Demand or Second Request, obtain the full current instructions and compare them against any planned or existing TAR/AI protocol. Assess whether your preferred tools and workflows can meet the 75 percent precision and recall floors and the pre-training disclosure timeline. Decide whether to negotiate specific modifications with FTC staff, prepare a statement of reasons for any non-compliant elements, or shift smaller collections to traditional search-term and manual review. Flag the prompt-disclosure requirement for privilege review, because some parties have argued that prompts can reveal attorney work product. Update outside counsel and e-discovery vendors immediately so that timelines and budgets reflect the new constraints.
What should operators watch next?
Watch whether the FTC applies these instructions uniformly or continues to negotiate case-by-case modifications. Monitor any court challenges to the requirements in the context of substantial-compliance disputes. Also track parallel developments at the Department of Justice and in civil litigation regarding AI-assisted review and prompt disclosure. For companies that regularly face FTC investigations, the practical question is whether the new rules will push a lasting shift away from TAR/AI for FTC matters or simply lengthen the negotiation and validation phase.
Frequently asked questions
Do these requirements apply to every FTC subpoena?
They have appeared in recent Civil Investigative Demands and merger Second Requests. Confirm the exact instructions in any specific matter.
What are the new performance thresholds?
Recall of at least 75 percent (with a ±3 percent margin of error at 95 percent confidence) and precision of at least 75 percent, with a cutoff score no lower than 0.50.
Must collections be finished before training?
Yes. The instructions state that training may begin only once collection from all custodians is complete and the FTC has reviewed the required disclosures.
Are prompts and rationales discoverable under these rules?
The instructions require disclosure of prompts or review protocols provided to the AI, plus relevance scores and rationales if available. Privilege arguments may be raised, but the requirement itself is explicit.
What are the practical options if the requirements are burdensome?
Parties may negotiate modifications, revert to search-term and manual review for smaller sets, or proceed under a protocol they consider reasonable while providing a statement of reasons. The last approach carries risk of a substantial-compliance dispute.
Son Güncelleme / Last Updated: October 10, 2026.
Related: Law department hub · FTC closes Gildan Robinson-Patman inquiry · OpenAI Astra for Law
Discover more from Kurums | Business Intelligence
Subscribe to get the latest posts sent to your email.
