Stanford's Default Assistant cuts legal review errors by 62% and time by 34%
AI tool boosts accuracy 6% and speed 26% in default judgment audits.
Researchers from Stanford (Theodora Worledge, Othman Bensouda Koraichi, Daniel Bernal, Aviv Caspi, Tatsunori Hashimoto, Carlos Guestrin, David Freeman Engstrom) developed the Default Assistant, an AI system that uses large language models to evaluate default judgment cases against legal requirements and provide cited recommendations. Their audit of 188 debt collection cases from Los Angeles Superior Court revealed troubling stats: 4% contained major defects that should have prevented judgment entirely, 10% had inconsistencies requiring reduced judgments, and 32% had errors needing amendment before judgment.
In a controlled study with 66 law students simulating court review (with more time and resources than actual court staff), the Default Assistant delivered significant gains. Users aided by the tool were 6.0% more accurate on average requirements (p < 1.0e-4) and 25.9% faster (p < 2.5e-10). The largest improvements came from statutory requirements demanding extensive document search, where AI assistance reduced errors by 62% and cut review time by 34% relative to unaided users (p < 0.05). The system grounds its outputs in cited quotes and tables from original case filings, enabling users to verify recommendations.
- Audit of 188 LA County default judgments found 4% had major defects, 10% inconsistencies, and 32% errors
- LLM-powered Default Assistant improved reviewer accuracy by 6.0% and speed by 25.9% in a 66-person study
- For complex document searches, AI cut errors by 62% and reduced review time by 34%
Why It Matters
AI assistants with citations can help resource-strapped courts review millions of default judgments more accurately and efficiently.