OpenAI Launches Astra for Law, a GPT-6 Astra Configuration Built for Legal Professionals

OpenAI has announced Astra for Law, a legal-focused configuration of its flagship GPT-6 Astra model, designed to support professional legal work ranging from case research and argument development to drafting advice and structuring deals. The announcement, made around 17 September 2026, positions the product not as a replacement for lawyers but as a tool to support their expertise and judgement.

The headline figure is the scale of the Legal Search Index bundled with the product: more than 230 million URLs covering U.S. Case law, statutes, regulations, court rules and administrative decisions, updated on a regular basis. That’s a sizeable corpus, and it’s what separates Astra for Law from simply pointing GPT-6 Astra at the open web.

What Astra for Law Actually Is

It’s worth being precise here: Astra for Law is not a new, separately trained model. According to independent technical coverage, it is a configuration of GPT-6 Astra — the same base model weights — with a dedicated legal index, custom instructions for legal analysis and writing, and specific settings layered on top. Think of it less as a specialist AI and more as GPT-6 Astra wearing a barrister’s wig, equipped with a very large law library and told to behave accordingly.

Access is initially restricted through a Trusted Access programme within ChatGPT and Codex, available to selected law firms and legal technology companies. An API version looks set to follow under the model identifier gpt-6-astra-law, appearing to users as “GPT-6 Astra Law.” Legal technology firms Harvey and Legora are among those named as prospective API customers. Specific rollout dates for UK firms have not been confirmed.

The Benchmark Numbers — and Their Limits

External analysis, citing results from the Vals AI Legal Research Bench — a private validation set of 200 U.S. Legal research questions — reports that Astra for Law achieved 54.0% overall correctness, compared with 38.7% for GPT-6 Astra using web search alone. That’s a meaningful gap, suggesting the legal index and custom instructions do add genuine value over the base model in a raw research context.

But these figures come from a private benchmark, not a publicly verifiable dataset. They should be treated as indicative rather than definitive. A 54% correctness rate also means nearly half of answers on that test set were wrong — a figure that matters enormously in high-stakes legal work, where an erroneous citation or a missed precedent can have serious consequences for a client.

OpenAI’s own communications acknowledge this directly. The company states that lawyers should review sources and outputs themselves, framing Astra for Law as AI-assisted legal work rather than AI-automated legal work. That’s a meaningful distinction, and one that regulators are likely to scrutinise closely.

Professional and Regulatory Questions

Access is restricted to eligible lawyers and people working under their supervision. That gatekeeping reflects the professional sensitivity around legal advice — in England and Wales, the Solicitors Regulation Authority and Bar Standards Board impose obligations on solicitors and barristers around competence, supervision, and client confidentiality. No Astra-specific guidance from either regulator has yet been issued, and any formal UK regulatory response remains unverified at this stage.

The EU AI Act and ongoing UK government work on AI governance add further regulatory context. Legal technology commentators have raised questions about data protection, legal professional privilege, and the risk that firms under competitive pressure might over-rely on AI outputs without adequate checking. Some describe the product launch as another move in the race for law firm AI adoption, with commercial pressure potentially outpacing settled policy.

OpenAI’s official position is that Astra for Law provides, in the company’s words, “frontier intelligence built for your practice,” with controls on who can use it designed to align with professional expectations.

U.S. Focus Raises Questions for Non-American Practitioners

The Legal Search Index at launch covers U.S. Legal materials only. That’s a significant constraint for law firms outside the United States, and raises the question of whether outputs could be misleading when applied to UK, EU, or other jurisdictions — chiefly in cross-border matters where practitioners might be tempted to draw on the tool’s apparent confidence without recognising the jurisdictional limits.

Integration of UK legislation, English and Welsh case law, or Scottish law into Astra for Law has not been confirmed. Any expansion of the index to cover non-U.S. Sources remains unverified.

That said, large law firms handling international matters — including those with transatlantic practices — may find the U.S. Legal index genuinely useful for the American side of cross-border work. Legal technology companies like Harvey and Legora are likely to build on the API to create more jurisdiction-specific products over time.

What This Means for Kent Residents

Astra for Law is not a consumer product, and there’s no evidence of any immediate deployment by Kent-based law firms. Early access is focused on larger, mainly U.S. Firms and leading legal technology companies. Over time, if solicitors and legal advisers serving Kent clients adopt tools like Astra for Law, residents might see changes in how quickly legal research is completed and, potentially, what that work costs — though no pricing has been publicly confirmed in GBP or any other currency. For now, anyone in Kent needing legal help should continue to go through their solicitor, barrister, or local law centre, as they always have.