AI Firms Race to Build Specialized Tools for Legal Work
The demand for artificial intelligence systems specifically designed for legal work has increased as law firms and legal technology companies seek tools capable of managing research and various professional tasks. In his 2023 year-end report, US Chief Justice John Roberts cautioned against the potential dangers posed by generative AI in the legal field, highlighting the likelihood of generating inaccurate or fabricated information. He emphasised that the application of such technology necessitates ‘caution and humility’. Since then, AI companies and legal technology providers have persistently advanced systems tailored for legal workflows. Last week, OpenAI broadened its enterprise AI initiatives within the legal sector by introducing Astra for Law, a tailored iteration of its GPT-6 Astra model specifically designed for the industry. The development arose as reports indicated that Anthropic is contemplating a new model in response to Astra’s initial success with enterprise users. Earlier this month, Thomson Reuters announced the development of its own proprietary large language model aimed at legal and compliance work.
OpenAI’s Astra for Law will first be accessible to a select group of law firms via its ‘Trusted Access’ program, with plans to subsequently extend availability through its application programming interface. Thomson Reuters is currently integrating its model into the workflow of CoCounsel Legal’s Tabular Analysis, an AI-driven document review tool. OpenAI unveiled Astra for Law on September 17, establishing a new foundation for law firms and legal technology enterprises. The model integrates GPT-6 Astra with tailored legal settings, specialised tools, contextual elements, and a legal search index. In a recent blog post, OpenAI announced that API customers such as Harvey and Legora will have the opportunity to develop products and workflows utilising the model. The company is also providing 26 ecosystem plugins that integrate ChatGPT with specialised tools utilised by law firms, such as Relativity and Clio. The legal search system is structured to locate US case law, statutes, regulations, court rules, and administrative decisions. “The index covers more than 230 million URLs, with sources added daily. Its work with Free Law Project also gives the system access to a case-law collection covering more than 99.9 per cent of published US precedential case law,” OpenAI said.
OpenAI conducted an evaluation of Astra for Law using 200 US legal research questions sourced from the private validation set of Vals AI’s Legal Research Bench. At the highest reasoning setting, OpenAI reported that Astra for Law achieved an overall correctness rate of 54 percent on the benchmark’s questions, in contrast to 38.7 percent for GPT-6 Astra utilising web search exclusively. OpenAI characterised this as a 40 percent relative enhancement. On case-law questions, OpenAI reported that Astra for Law identified 24 percent more reference cases compared to GPT-6 Astra utilising web search alone. In an audited set of target passages, it successfully retrieved up to 54 percent more relevant passages from the appropriate court opinions. “The system is also configured for legal analysis and writing. It can work with client facts, develop arguments or deal terms, and identify weaknesses and areas of uncertainty,” OpenAI said. The company stated that Astra for Law can, for instance, differentiate a court’s holding from other observations and pinpoint cases that undermine an argument. Anthropic is contemplating the introduction of a new AI model prior to its IPO, responding to competitive pressures from OpenAI’s GPT-6 Astra, as reported.
The company has not yet made a final decision regarding the release. Anthropic is also assessing the safety of its forthcoming model in the context of deliberations regarding a potential release, the report noted. The company is reportedly evaluating the investment in new models in relation to profitability, as investors are increasingly attentive to the timeline for AI companies to achieve sustainable profits. OpenAI unveiled GPT-6 Astra on September 3, emphasising its applications in computer use, software engineering, cybersecurity, and professional work. Source reported that the model has garnered significant interest from enterprise users. Data indicates that Astra represents approximately 13 percent of enterprise AI spending, while Anthropic’s Claude Fable accounts for around 8 percent. According to the platform, OpenAI models outperformed Anthropic’s models in terms of user spending on OpenRouter last week. Source reported that this marked the first occasion OpenAI had taken the lead on that measure in over two and a half years. Anthropic continues to exhibit a greater reported annualised revenue run rate. Source reported that it surpassed $65 billion by the end of July, in contrast to OpenAI’s figure of over $40 billion. OpenAI and Anthropic are not the sole entities engaged in the development of specialised AI for legal applications.
Thomson Reuters has developed Thomson, its proprietary LLM designed for high-stakes professional tasks, encompassing legal and compliance applications. Thomson Reuters said the model uses its proprietary content, including Westlaw and Practical Law, alongside input from its network of legal experts. The company stated that possessing a larger share of the AI stack provides enhanced control over performance, economics, deployment, and data protection. The company said its decision to build the model began in 2024 after it acquired Safe Sign Technologies, an AI research company founded by AI and legal experts. Thomson is built on open-source foundations that Thomson Reuters said it can continue to evolve. Thomson Reuters announced that the initial application of the model supports CoCounsel Legal’s Tabular Analysis feature. The tool can extract structured answers from up to 10,000 documents across up to 100 questions, with citation-grounded outputs. The company said CoCounsel had crossed one million users as of the first quarter of 2026, while the share of its business that is generative-AI enabled had doubled in a little over a year.
