OpenAI Enters the Courtroom: Inside Astra for Law

Artificial intelligence has been creeping into legal practice for years, mostly around the edges: document review, contract summaries, e-discovery. OpenAI's newest release pushes considerably further. 

Astra for Law is a purpose-built system aimed at the core intellectual work of lawyering, from research and analysis to drafting, and it is being offered to law firms and legal technology companies as a foundation they can build on.

• What the System Actually Is

Astra for Law is not simply a chatbot with a legal coat of paint. It pairs GPT-6 Astra with a dedicated legal search index, a set of research tools, and instructions written specifically for professional legal work. The intention, according to OpenAI, is to give firms and vendors a base layer on which they can construct their own applications and internal workflows rather than a finished product handed down from above.

Read Also: OpenAI Enters Browser Market With ChatGPT Atlas

The search index is the headline component. It spans more than 230 million URLs, taking in United States case law, statutes, regulations, court rules and administrative decisions, with new sources added daily. To deepen the case-law coverage, OpenAI worked with the Free Law Project, the non-profit behind the CourtListener database, bringing in a collection that covers over 99.9% of published US precedential case law.

The workflow OpenAI describes will sound familiar to any practising lawyer. You begin with the facts of a matter. The system helps you move from those facts to the authorities that govern them, surfacing the specific passages that matter. Its legal instructions then assist in applying those authorities to the client's circumstances, building arguments, probing for weaknesses and flagging where the law is genuinely uncertain rather than papering over the gaps.

• The Numbers Behind the Claims

OpenAI evaluated the system against 200 US legal research questions drawn from the private validation set of Vals AI's Legal Research Bench.

At its highest reasoning setting, Astra for Law passed the benchmark's overall correctness check on 54% of questions. GPT-6 Astra relying on ordinary web search managed 38.7%. On case-law questions specifically, the specialised system surfaced 24% more reference cases, and on an audited set of target passages it retrieved up to 54% more relevant material from the correct court opinions.

Those are meaningful gains, but they deserve context. The testing was conducted by OpenAI itself, and it focused squarely on US legal research. Nothing in the results speaks to performance in other jurisdictions or across the wider range of tasks lawyers actually perform. A 54% correctness rate also means that nearly half the answers fell short of the mark, which is a useful reminder that this is a research assistant, not an oracle.

• Firms Are Already Building

Several prominent firms have been working with OpenAI on applications shaped around their own expertise and internal processes, which is arguably the more interesting story.

Sullivan & Cromwell built an agreement analyser that draws on the firm's negotiating playbooks and selected precedents to identify risks, propose redlines and prepare guidance for clients. Ropes & Gray developed a due diligence system that reviews data-room materials and flags issues relevant to a transaction. Cooley created GO Public, aimed at companies preparing for an initial public offering.

The pattern is worth noting. In each case the firm's proprietary knowledge, accumulated over decades, remains the differentiator. The AI supplies speed and reach; the institutional judgement still comes from the firm.

Legal technology companies are in the mix too. Harvey and Legora will be able to build products on Astra for Law through OpenAI's API, and 26 partner-built plugins are launching alongside it, with integrations spanning Relativity, Clio, iManage, Intapp, DeepJudge and Thomson Reuters.

• Access Is Deliberately Narrow

For now, Astra for Law is available only to selected firms through OpenAI's Trusted Access programme. That gating reflects the obvious sensitivity of legal data.

Eligible firms get Zero Data Retention on the API, and ChatGPT Enterprise usage is excluded from human review by default. OpenAI is also working with Latham & Watkins on governance controls covering information permissions, ethical walls, client instructions and firm-level oversight, all of which speak to the confidentiality obligations that make law a difficult market for general-purpose AI tools.

Within ChatGPT, the system appears in the model selector as GPT-6 Astra Law. The API version is identified as gpt-6-astra-law. Broader API access is promised, though OpenAI has not committed to a date or published pricing.

• What It Signals

The strategic message is clear enough. Rather than competing head-on with established legal technology vendors, OpenAI is positioning itself as the infrastructure those vendors and firms build upon. Harvey and Legora become customers, not casualties.

For the profession, the implications are more gradual than dramatic. Research that once consumed junior associates for days may compress into hours. That changes the economics of training, billing and staffing in ways firms will spend years working out.

For now, though, the verification burden stays exactly where it has always been: with the lawyer who signs the brief.