OpenAI Astra for Law Release

Astra for Law provides a specialized foundation for legal AI

OpenAI has launched Astra for Law, a system designed for law firms and legal technology companies to build AI products and workflows. The offering combines the GPT-6 Astra model with specialized settings, tools, and a legal search index tailored for professional legal work. It is available to selected law firms via Trusted Access in ChatGPT and Codex, and will be released to the API as gpt-6-astra-law.

Enhanced legal research capabilities

Astra for Law integrates a legal search index covering more than 230 million URLs, including U.S. case law, statutes, regulations, court rules, and administrative decisions. This index incorporates the case-law collection from the Free Law Project (CourtListener), which covers over 99.9% of published U.S. precedential case law.

Performance benchmarks on Vals AI Legal Research Bench

Testing on 200 U.S. legal research questions from the Vals AI Legal Research Bench private validation set demonstrated significant improvements over standard GPT-6 Astra using web search:

  • Overall Correctness: Astra for Law achieved a 54.0% correctness rate compared to 38.7% for GPT-6 Astra (a 40% relative improvement) at the highest reasoning effort.
  • Case Law Retrieval: Astra for Law found 24% more reference cases than GPT-6 Astra using web search alone.
  • Passage Retrieval: The system retrieved up to 54% more relevant passages from correct court opinions when compared at the same reasoning effort.

End-to-end legal workflow optimization

Astra for Law uses custom instructions for legal analysis and writing to help practitioners apply research to client facts, develop arguments, and identify legal uncertainties. This includes distinguishing court holdings from observations and addressing cases that weaken a specific argument.

In a comparative test regarding a misrepresentation claim, Astra for Law successfully pinpointed two closely matching precedents, whereas Claude Fable 5.1 returned a holding that had been reversed on appeal and failed to find a case in a transactional example.

Legal-grade privacy and governance

To meet the confidentiality requirements of legal practice, OpenAI introduced a Trusted Access Program for eligible firms. Key security features include:

  • Zero Data Retention (ZDR): Available on the API for eligible firms.
  • Human Review Exclusion: ChatGPT Enterprise usage is excluded from human review by default for eligible firms.
  • Governance Partnership: OpenAI is collaborating with Latham & Watkins to design information permissions, ethical walls, and firm oversight mechanisms.

Firm-specific AI implementations

OpenAI's forward-deployed engineers have worked with several firms to create custom interfaces and integrations for proprietary data:

  • Sullivan & Cromwell: Developed an agreement analyzer using negotiating playbooks and precedents to spot risks and propose redlines.
  • Ropes & Gray: Created a deal diligence system to trace findings back to sources within data rooms.
  • Cooley: Built "GO Public" to assist with IPO filings and risk identification.

Ecosystem integration and plugins

OpenAI has released 26 partner-built plugins to connect ChatGPT to existing legal tools. These include integrations with:

  • iManage: For drafting and saving negotiation briefs to matter files.
  • Intapp: For surfacing activities requiring time entries.
  • DeepJudge: For comparing prior deals.
  • Thomson Reuters: Integrating HighQ matter context and previewing a CoCounsel Legal connector.

Additionally, the launch includes 9 community plugins from LegalQuants, LECG, and Skills.law, and 47 custom skills. ChatGPT for Word is also now generally available to assist with proofreading and formatting in legal drafting.

Sources