OpenAI launches Astra for Law with a legal search index

OpenAI said today it is introducing Astra for Law — GPT-6 Astra configured as a foundation for law firms and legal tech — with a dedicated legal search index, legal analysis/writing instructions, and expanded privacy controls for confidential client work.

What shipped: Astra for Law combines GPT-6 Astra with a legal search index covering U.S. case law, statutes, regulations, court rules, and administrative decisions across 230M+ URLs (sources added daily), built with Free Law Project / CourtListener. OpenAI says the collection covers more than 99.9% of published U.S. precedential case law. On the private validation set of Vals AI’s Legal Research Bench (200 questions), at highest reasoning effort, Astra for Law scored 54.0% overall correctness vs 38.7% for GPT-6 Astra with web search alone (~40% relative improvement). On case-law-focused questions it found 24% more reference cases; on audited target passages, up to 54% more relevant passages — company-reported, not an independent audit we re-ran. API customers including Harvey and Legora can build on it. 26 new ecosystem plugins connect ChatGPT to specialist legal tools (OpenAI cites Relativity and Clio as examples). Initially offered to selected law firms through Trusted Access in ChatGPT and Codex; coming soon to the API. Model picker name: “GPT-6 Astra Law”; API id: gpt-6-astra-law. Eligible firms under a Trusted Access Program get Zero Data Retention on the API, and ChatGPT Enterprise usage excluded from human review by default, per OpenAI.

What it cannot do: This is not a general public launch — selected firms via Trusted Access first; API still “coming soon.” OpenAI’s post does not publish pricing, retention SLAs beyond the ZDR/Enterprise defaults named above, or third-party audits of the Vals numbers. A legal search index is not a substitute for licensed research products firms already buy (OpenAI says it complements providers such as Thomson Reuters).

Who it is for: Selected law firms in Trusted Access; legal-tech builders already on the OpenAI API (Harvey, Legora named); firms that want ChatGPT connected to tools they already run via the new plugins.

Risk: Company-reported bench gains can overstate courtroom or deal-desk reliability. Confidential-client controls (ZDR, no default human review on Enterprise) are contract/product claims — clients will still need firm-side policy and vendor diligence. Grounding in “authority” does not remove hallucination or citation-error risk on matters outside the indexed corpus.

What’s next: Broader Trusted Access; API GA for gpt-6-astra-law; any independent replication of the Vals Legal Research Bench results; regulator or bar-association scrutiny of AI legal research marketed to clients.

Harvey’s Niko Grupen (Head of Applied Research), in OpenAI’s post: “In our early testing, Astra for Law showed strength across key aspects of legal research: grounding answers in on-point authorities, citing with precision, and offering practical, advisory guidance.”

— Rowan Keel