The legal AI market is drawing in major technology vendors even as courts keep penalizing lawyers who file AI-generated material containing fake case citations. The trend spans from Google and Anthropic, which have built products that steer the model toward outside legal databases, to SpaceXAI, whose Grok-powered legal offering has published less detail about how it verifies citations. The stakes fall on US and Canadian attorneys, who remain responsible for anything they submit regardless of the tool used to produce it, and on the courts now sorting through the resulting errors.
The enforcement record is already substantial. In June, the Ontario Law Society Tribunal ordered lawyer Shahryar Mazaheri to pay $31,150 after finding he used an earlier version of Grok to draft a factum in a client's appeal, producing fabricated citations and invented legal principles. The adjudicators said the LLM lacks nuance and judgment and tends to produce an answer rather than admit ignorance, describing the AI-generated materials as gibberish. Researcher Damien Charlotin's database has cataloged nearly 2,000 such cases worldwide, involving more than 800 lawyers and 1,100 self-represented litigants.
Large firms are among those caught. Sullivan & Cromwell filed an emergency letter with the Southern District of New York in April after a bankruptcy motion submitted to Chief Judge Martin Glenn contained 42 AI-generated fabrications. In Canadian courts alone, reported cases involving fabricated citations rose from seven in 2024 to 86 in 2025, with 39 more in the first quarter of 2026. Earlier, DoNotPay dropped plans in early 2023 to have an AI robot lawyer argue a case after state bar associations threatened criminal prosecution.
Google and Anthropic say their legal tools are designed for verification. Google routes queries through connectors to external legal databases including Everlaw and NetDocuments rather than relying on the model's training data alone, and Google Cloud CEO Thomas Kurian said at launch that grounding agentic workflows in legal authority is critically important. Gottlieb, Freshfields, Weil and Williams & Connolly are in the preview program. Anthropic's Claude Legal Solutions includes a reported 20 connectors to legal platforms and 12 pre-built plugins, claims a 90.9 percent score on the BigLaw Bench legal reasoning benchmark for Opus 4.7, and counts Freshfields and Quinn Emanuel as customers.
Grok's legal positioning is less developed. A Cursor blog post said Grok 4.5 suited finance, legal work or other computer tasks, but SpaceXAI's later announcement for the model discussed coding, agentic tasks and knowledge work without mentioning legal applications, and the page discloses no architecture or mechanism for tracing citations to primary sources. SpaceX acquired Cursor for $60 billion on August 15. A 2024 Stanford RegLab study found hallucination rates of 17 percent and 33 percent in purpose-built legal AI platforms from LexisNexis and Westlaw, though no comparable independent audit has examined Google or Anthropic's legal tools.
Bar association rules place the duty to verify citations on the filing attorney whether or not AI was used. Mazaheri's penalty followed from output that was not thoroughly checked, not from his use of AI itself.
More AI news from TechManNews.
