OpenAI launches Astra for Law for legal firms

OpenAI has introduced Astra for Law, a version of its AI technology built for legal work. It combines the GPT-6 Astra model with legal search, instructions for analyzing and drafting documents, specialized tools, and privacy controls designed for confidential client information.

The offering is not aimed only at lawyers who want to use a chatbot. It also allows firms and legal technology companies to build their own applications and workflows around their expertise, precedents, and internal methods.

AI that searches legal sources

Astra for Law includes a legal search index that queries US case law, statutes, regulations, procedural rules, and administrative decisions. The corpus contains more than 230 million URLs and receives new sources every day.

OpenAI has also worked with Free Law Project, the organization behind CourtListener, to include a collection covering more than 99.9% of published and binding US case law. The tool is designed to locate the right authority, find the relevant passages, and assess how they relate to a case.

This complements, rather than replaces, licensed-content databases and specialized products that firms already use, such as those from Thomson Reuters.

Better results in a research test

OpenAI tested the full Astra for Law configuration with 200 US legal questions from the private validation set of Legal Research Bench, created by Vals AI. In the evaluation, the system had to find relevant sources and passages and meet the established accuracy criteria.

At the highest reasoning level, Astra for Law passed the overall accuracy review on 54% of the questions, compared with 38.7% for GPT-6 Astra using web search alone. That represents a relative improvement of 40%.

On questions focused on case law, it found 24% more reference cases. In addition, on the audited set of target passages, it retrieved up to 54% more relevant excerpts from the correct judicial opinions, always compared with the same model using web search and the same reasoning effort.

These figures do not mean the tool can replace a lawyer's review. They do point to a practical advantage: reducing the time needed to find authorities and making it easier for professionals to verify the source before using it.

From finding a document to developing a strategy

Search is only the first step. Astra for Law's specialized instructions help apply research to a client's facts, develop arguments, draft contract terms, and identify weak points or uncertainties.

For example, it can distinguish between a court's central holding and secondary commentary, search for precedents that undermine a position, or explain how a contractual exception allocates risk between the parties.

OpenAI is already working with firms to create tools adapted to their processes:

• Sullivan & Cromwell developed an agreement analyzer that uses its negotiation manuals and precedents to detect risks across clauses, suggest changes, and prepare draft advice.
• Ropes & Gray created a diligence system to review a data room and identify relevant issues in an acquisition, such as contracts requiring notices or consents.
• Cooley developed GO Public to support IPO preparation, from the registration statement to identifying risks for management.

Controls for confidential information

Legal firms need to control who can see each document, how AI is used, and what happens to client data. That is why the Trusted Access program for eligible firms includes zero data retention in the API and excludes ChatGPT Enterprise conversations from human review by default.

OpenAI is also working with Latham & Watkins on information permissions, ethical walls between teams, client instructions, and internal oversight. Astra for Law will initially be offered to selected firms through Trusted Access in ChatGPT and Codex, and will soon come to the API as gpt-6-astra-law. In ChatGPT, it will appear as GPT-6 Astra Law.

An ecosystem of legal tools

The launch includes 26 plugins created by partners to connect ChatGPT with tools firms already use. Among them, iManage can save a negotiation report to the matter file; Intapp can detect activities that should be converted into time records; and DeepJudge can compare previous transactions.

Thomson Reuters will provide matter context from HighQ and is preparing a CoCounsel Legal connector. Nine community plugins from LegalQuants, LECG, and Skills.law are also arriving, with 47 customizable skills. In addition, ChatGPT for Word is becoming generally available for reviewing text, suggesting edits, and detecting formatting issues.

For you, the most important change is not having another assistant that drafts contracts. It is that firms can start turning their own precedents, standards, and processes into AI tools with controls they define themselves. The next thing to watch will be whether access to better sources translates into fewer errors and legal decisions that are easier to review, not just documents produced faster.