OpenAI has unveiled "Astra for Law," an AI platform designed for legal work. It combines GPT-6 Astra with a dedicated search index for U.S. legal materials and guidelines for legal analysis and drafting, enabling law firms and legal tech companies to create specialized workflows.
It will be listed as "GPT-6 Astra Law" on the model selection screen in ChatGPT and Codex. However, it was announced that at the time of launch, only select U.S. law firms will be able to access it through the Trusted Access program, and the API will be available soon.
Therefore, it would be inaccurate to view this as the general release of a completely new, standalone model. It is a service built around GPT-6 Astra, the latest general-purpose model, configured for legal research and with dedicated settings; however, its scope of use is still limited.
| Integrating the legal work environment with a general-purpose model
Astra for Law isn't just about GPT-6 Astra on its own. It bundles legal research tools, analytical guidance, and the ability to connect with law firm data and specialized software to provide a platform tailored to real-world legal work.
In the legal field, it’s more important to accurately identify which case law and statutes apply—and to verify that those sources are still valid—than to craft well-written sentences. That's why a dedicated search index became a key tool for finding the basis for answers.
The tailored guidelines distinguish between the core findings of a judgment and ancillary opinions, and encourage you to examine even the risks posed by case law or contract provisions that are unfavorable to your argument. Even so, Since it is not a system that automatically guarantees the legal accuracy of the results, expert review is still necessary.
| A search index covering more than 230 million URLs
According to OpenAI, its dedicated legal search index covers more than 230 million URLs, including U.S. case law, federal and state laws, regulations, court rules, and administrative decisions. The company explained that new data is added daily.
The CourtListener database—which covers more than 99.9% of publicly available U.S. case law and was made available through a partnership with the nonprofit organization Free Law Project—has also been integrated into the search experience. While this helps users find a wide range of public legal resources, That doesn't mean it will replace everything, from paid legal databases to a law firm's internal knowledge.
OpenAI also explained that this index complements licensed content and specialized products from providers such as Thomson Reuters. This is why it is still necessary to verify the jurisdiction, currency, and citation status of legal materials.
| Search accuracy improved in our internal evaluation
OpenAI announced that it tested Vals AI’s Legal Research Bench on a private validation set of 200 U.S. legal research questions. At the highest level of reasoning, Astra for Law achieved an overall accuracy of 54.0%, while GPT-6 Astra, when using general web search, achieved 38.7%.
The company described this as a 40% improvement compared to previous results. The study found that, for case law-based questions, participants identified 24% more relevant precedents and, at the same level of reasoning, retrieved up to 54% more relevant paragraphs from correct court opinions.
However, there is a limitation in that it is a closed verification set of 200 questions. Since the results were obtained under settings and evaluation conditions chosen by the company, Further independent verification is needed to determine whether the system delivers the same performance across the various jurisdictions and complex cases encountered in real-world law firms.
| Initially, only certain U.S. law firms were used
Astra for Law isn't a service open to the general public from the start. Selected U.S. law firms will be the first to use it through the ChatGPT and Codex Trusted Access programs.
OpenAI announced that it has tested and developed use cases for legal AI in collaboration with firms such as Sullivan & Cromwell, Ropes & Gray, Cooley, Latham & Watkins, and Wachtell Lipton. This appears to be an effort to focus on validating specialized workflows—such as contract analysis, legal research, and drafting legal opinions—during the initial stages of adoption.
The API's model identifier was announced as "gpt-6-astra-law," but at the time of its release, it was listed as "coming soon." It was also announced that legal AI companies such as Harvey and Legora will be able to build their own products and workflows based on APIs in the future.
| Connecting law firm data with existing specialized tools
OpenAI also introduced 26 ecosystem plugins that connect ChatGPT to specialized tools already used by law firms, such as Relativity and Clio. This strategy aims to integrate AI into law firms’ existing data and workflows rather than simply providing a generic chat interface.
Selected law firms can develop their own tools by leveraging internal case law, negotiation principles, and review criteria. For example, they can create a workflow that analyzes the risks associated with various clauses in a new contract and proposes amendments and draft explanations for clients.
While such connections can boost productivity, if access rights aren't properly managed, there is a risk that other customers' confidential data could be mixed in or exposed more than necessary. The classification of data by law firm, access logs, and responsibility for reviewing results must be managed through a combination of technical settings and internal regulations.
| Confidentiality and legal liability remain key challenges
OpenAI has proposed safeguards for participating law firms, including the API’s “Zero Data Retention” policy and the exclusion of basic human review of ChatGPT Enterprise usage history. The company stated that it is working with Latham & Watkins to design information access controls, ethical firewalls, client guidelines, and a law firm oversight structure.
However, the mere fact that data isn’t stored doesn’t mean all risks in legal work disappear. Issues such as citing the wrong case law, confusing jurisdictions, improperly sharing confidential documents, and placing excessive trust in AI results still remain.
The responsibility for the final legal determination and the documents submitted lies with the attorney and the law firm. That’s why Astra for Law should be used as a tool to expand the scope of search and drafting rather than as an automated decision-making tool that replaces legal professionals.
| The Next Benchmark in the Legal AI Race
The significance of this release lies not only in the launch of a larger language model. It combines a search system for finding reliable, specialized resources with law firm workflow tools and confidentiality controls into a single environment.
At this point, the only things that have been finalized are the limited offering to select law firms, the product structure, and the evaluation results. The specific dates for the public launch and API availability, the actual pricing, and the schedule for expanding the supported jurisdictions have not yet been finalized.
Going forward, we need to verify whether citation accuracy is consistently maintained in independent evaluations, whether outdated or superseded case law is properly distinguished, and whether the law firm’s confidentiality controls are actually upheld in day-to-day operations. It is highly likely that competition in legal AI will shift from simple answer-generation capabilities to a focus on the authority and traceability of data, as well as structures that ensure responsible use.
Tags
#OpenAI #AstraForLaw #GPT6AstraLaw #Legal AI #LegalTech #Legal Search #Generative AI #ChatGPT #Codex #Harvey #Legora #Law Firm #AI Governance
