Trust, risk & security
Risk & Hallucination Rates
Estimated error rates across legal tasks, highest for citation and jurisdiction work.
As a directional estimate, error and hallucination risk varies widely by task type rather than applying as a flat rate across legal AI use. Document summarization sits at the low end (1.8%), while tasks requiring precise sourcing or cross-jurisdictional reasoning score highest: litigation strategy at 8.5%, jurisdiction analysis at 7.3%, and case law citation at 6.2%. The pattern suggests risk concentrates where a model must anchor claims to specific external authorities rather than restate or synthesize provided text, reinforcing the case for citation-verification layers on higher-stakes research and drafting workflows.
The data
| Task | Error rate |
|---|---|
| Legal Research | 3.8% |
| Contract Drafting | 2.1% |
| Compliance Analysis | 4.5% |
| Case Law Citation | 6.2% |
| Statutory Interpretation | 5.1% |
| Due Diligence | 3.4% |
| Document Summarization | 1.8% |
| Client Communication | 2.9% |
| Jurisdiction Analysis | 7.3% |
| Regulatory Filing | 4.8% |
| Litigation Strategy | 8.5% |
| Tax Advisory | 5.7% |
Illustrative estimate - a directional figure for scenario framing, not a measured benchmark. Do not read these as measured per-vendor results.
Related research
- Client Trust & SatisfactionTrust, satisfaction and reuse rates for AI across legal work products.
- Cross-Model Legal BenchmarkHAQQ's own measured benchmark scoring legal AI models out of 50 across legal task categories.
- Data Security & CompliancePublicly documented security certifications across major AI providers.