OpenAI is launching Astra for Law, a specialized version of GPT-6 Astra built specifically for legal work. It bundles the base model with a legal search index and custom instructions designed for legal research, analysis, and writing. The system aims to handle the full workflow—finding relevant case law and authorities, then applying that research to specific client situations by developing arguments, spotting weaknesses, and identifying deal terms. Early testers from Sullivan & Cromwell praised its ability to ground answers in relevant authorities and cite with precision, which matters because accuracy and sourcing are non-negotiable in legal work. The product isn't meant to replace existing legal research tools from Thomson Reuters or other established providers. Instead, it's positioned as a foundation that law firms and legal tech companies can customize and build on. This matters because every firm has different expertise areas and client bases—a corporate M&A shop needs different workflows than a litigation practice. OpenAI is working with Latham & Watkins on governance features like information permissions and ethical walls, which addresses a real pain point: AI tools handling sensitive client data need built-in safeguards that respect confidentiality and conflict rules. The underlying bet is that law firms want AI they can adapt to their own workflows rather than adopt an off-the-shelf product. By providing the model plus legal-specific training and tools, OpenAI is betting firms will layer this into their existing stacks—using it alongside LexisNexis or Westlaw rather than replacing them. The emphasis on traceability and source attribution suggests OpenAI understands that lawyers can't rely on a black box; they need to verify what the AI found and justify recommendations to clients.