Harvey uses GPT-6 Astra for legal drafting

Legal AI company Harvey is integrating GPT-6 Astra into its drafting workflows, aiming to help lawyers produce more structured and context-aware documents. According to the announcement, the system uses legal context—such as case materials, firm precedents, and matter-specific information—to generate stronger first drafts. The goal is not to replace legal judgment but to reduce the time lawyers spend on formatting, initial writing, and repetitive revisions, freeing them to focus on strategy and client advice. Law is one of the most document-heavy professions, and drafting quality depends on more than fluent prose. Legal writing must be accurate, traceable, and grounded in the specific facts and authorities of a matter. Generic language models can produce convincing text, but they often lack the context and reliability that legal work demands. Harvey's use of GPT-6 Astra appears designed to address that gap by anchoring output in case-specific material rather than relying on generic patterns. If it works as intended, the integration could cut down the back-and-forth that plagues many AI drafting tools. Lawyers typically need to fact-check, restructure, and rewrite machine-generated text, which can erase the promised time savings. A system that better understands the matter's context could produce drafts that require less rework, though human review remains essential. Legal ethics rules place responsibility for accuracy and confidentiality on lawyers, not software. The collaboration also highlights how quickly frontier models are moving into specialized professional tools. Rather than using a general chatbot, firms increasingly want AI embedded in secure platforms that understand their documents and workflows. Harvey has positioned itself as a legal-specific layer on top of advanced models, and its choice of GPT-6 Astra suggests confidence in the model's ability to handle complex, context-rich tasks. The announcement did not provide detailed benchmarks or independent evaluations, so claims about improved drafting should be treated with appropriate caution. The real test will come from firms using the tool on live matters, where accuracy, privilege, and auditability matter. If Harvey and GPT-6 Astra can deliver reliable drafts while keeping lawyers in control, they could change how legal teams allocate their time. If not, the profession's skepticism will remain well founded.