DOJ market-researches AI legal assistant platforms
The RFI asks whether a platform can execute multi-step legal tasks without step-by-step attorney prompting, a capability very different from the legal research tools DOJ already uses.
TL;DR
The Justice Department posted an RFI Friday seeking commercially available AI legal assistant platforms that can analyze documents and conduct legal research autonomously. A pilot would involve 20 to 30 government users on synthetic or public data, hosted in a vendor's FedRAMP-authorized or commercially available cloud. DOJ already uses LexisNexis, Westlaw, and Bloomberg for legal research, per its 2025 AI use case inventory, but those are search and retrieval tools, not platforms that execute multi-step legal tasks without attorney prompting at each stage.
The RFI asks vendors whether their platforms can synthesize 5,000 to 100,000 documents per session, output authoritative sources, assist with drafting, and integrate with existing agency document management systems. DOJ also wants to evaluate usage statistics, output quality reviews, timing logs, and anonymous employee feedback surveys. The SAM.gov posting is explicit that this is market research, not a solicitation, no contract award will result.
The capabilities being researched go beyond what DOJ's current AI toolkit covers. The department already fields FedRAMP-authorized SaaS platforms for e-discovery (Everlaw) and multimedia analysis (Veritone aiWARE), and its 2025 AI use case inventory lists LexisNexis, Westlaw, and Bloomberg for legal research. But an AI legal assistant that chains together document analysis, regulatory review, FOIA processing, contract clause extraction, and memo drafting (autonomously, without prompting between steps) would be a different category of tool.
DOJ didn't say why it's looking now or what timeline it envisions. But the evaluation criteria suggest the department is interested in more than a proof-of-concept: timing logs, quality reviews, and usage analytics are the sorts of metrics you collect when you're thinking about whether the thing could scale past 30 users.
The cloud requirement gives vendors two paths, FedRAMP-authorized or commercially available. That's a wider aperture than a strict FedRAMP mandate, though any production deployment would almost certainly need authorization. For now, the door is open to platforms that can demonstrate the capabilities in a controlled environment without having completed the full certification process.
Published ·Deep Fathom