487 AI Errors in Court: What Litigators Need to Know About Document Review in 2026
This article is a draft in progress. Full content coming soon.
The 487 Number and What It Means
In 2025, U.S. courts recorded 487 documented instances of AI-generated errors or hallucinations in legal filings — more than ten times the number from 2024. This data comes from Artificial Authority, as reported in the SurePoint 2025 State of the Legal Industry Report.
81% of Firm Leaders Are Concerned About Reliability
While 63% of mid-sized firms have formally adopted generative AI for legal work, 81% of firm leaders report concern about the reliability of AI outputs (SurePoint 2025). The gap between adoption and trust is where risk accumulates.
The Sanctions Risk
Courts have already imposed severe penalties for AI-assisted filings containing fabricated citations. In DR Distributors, the court issued a $2.5 million sanctions order related to discovery misconduct. As AI tools become standard in legal workflows, the obligation to verify AI-generated content before filing becomes non-negotiable.
What Grounded AI Review Looks Like
Attested enforces 10 grounding rules on every AI document review. Findings must cite specific source text from the document. Timelines include only explicitly dated events. Entities are extracted only from named references. A mandatory hallucination self-check runs before any review is finalized. Every review includes a confidence score so attorneys know when to trust the output and when to verify.
How to Evaluate AI Tools for Evidence Work
When evaluating AI tools for litigation, ask: Does the tool cite its sources back to the original document? Does it include a confidence score? Can it distinguish between what the document says and what it infers? Does it run a self-check for hallucinated content? These are baseline requirements for any AI tool handling legal evidence.
Try Attested's AI Review
Upload any PDF or image to Attested, click Review, and see grounded AI analysis with source-cited findings, entity extraction, timeline, and a document quality confidence score. Ask Counsel AI follow-up questions grounded in the same review data.