Terms trackerLegal tech platforms

Harvey AI Terms of Service — Key Clauses & Change History

Harvey is the fastest-growing AI legal platform, backed by $300M+ in funding and used by Am Law 100 firms and major in-house teams. Its Platform Agreement, DPA, and AI Policy have expanded 4.5x since 2023 — growing from 5KB to 23KB+ — reflecting Harvey's maturation from startup to enterprise vendor. DriftPatrol tracks harvey.ai/legal/platform-agreement and its companion documents across 65+ documented change events.

Best-in-class: AI training prohibition extends to ALL sub-processors
Current terms at a glanceEffective Jan 9, 2026 · DPA updated Dec 26, 2025
Section What it says Why it matters
§11.8 AI Training Prohibition — extends to ALL sub-processors Harvey will not train AI on your content. This prohibition flows down to every sub-processor — not just Harvey itself. Strongest no-train clause among all five major legal tech platforms reviewed. Zero carve-outs for "de-identified aggregation" unlike Clio. Best in class
§7.4 Data Deletion — within 30 days of termination 30-day deletion window post-termination. Shorter than Clio's 90-day retrieval window — plan your data export before you cancel, not after. Request a deletion confirmation certificate for regulatory compliance. Note
§11.14 Mandatory Arbitration — JAMS (US) / ICC (EU) · No court path Mandatory arbitration in San Francisco (US) or London (EU). Three-arbitrator panel for disputes over $250K. No path to court for US customers. JAMS fees for three-arbitrator panels can run $50K–$200K+ — a material deterrent for mid-market firm disputes. Risk
§11.2 Sub-processor notice — 30 days advance, 15-day objection window Harvey must give 30 days' advance notice before adding a sub-processor. You have 15 days to object. Short window — if your team isn't monitoring, you'll miss it. DriftPatrol flags sub-processor notices the day they're published. Monitor
§10.2–3 Liability — 12mo fees or $250K floor; data breach doubles to 24mo / $500K Meaningful liability floor: greater of 12 months of fees or $250,000 for standard claims. Data breach carve-out doubles both figures. Significantly stronger than Clio (6 months, no floor) and Fastcase (effectively zero). Strong
§11.13 Governing Law — California (US) · England and Wales (EU/UK) US customers governed by California law; EU/UK customers by English law. Clean split — no ambiguity about jurisdiction. California's CCPA adds an extra layer of consumer protection for client data processed in California. Note
§11.6 Confidentiality — reasonable care, same as own sensitive information Standard reasonable care obligation. Not a "highest degree of care" clause — monitor for any future weakening to "industry standard" language, which courts have interpreted broadly in vendors' favor. Note
Documented changes65+ events · 5KB → 23KB+ since 2023
Jan 9, 2026 — Most recent Platform Agreement update
harvey.ai/legal/platform-agreement · EU AI Act compliance language added
+ EU AI Act (Regulation 2024/1689) compliance commitments incorporated + High-risk AI system classification obligations addressed for EU customers + Conformity assessment documentation provisions added flagged: EU AI Act, compliance, regulatory impact: EU AI Act applies to legal AI systems. Harvey's explicit compliance commitment is a positive signal — but the TOS language does not specify which risk tier Harvey's systems are classified under. Request the conformity assessment documentation for your records if you are an EU customer or serve EU clients.
Dec 26, 2025 — DPA + AI Policy simultaneous update
harvey.ai/legal/data-processing-addendum + harvey.ai/legal/ai-policy · Dual update
+ Canadian personal data provisions added to DPA (PIPEDA + Quebec Law 25 compliance) + Business improvement processing language clarified — AI Policy now distinguishes "product improvement" from "model training" + EU AI Act Regulation 2016/1689 compliance commitment added to AI Policy flagged: Canadian privacy law, AI policy, product improvement impact: The "business improvement processing" clarification matters. Harvey's AI Policy now explicitly distinguishes this from LLM training — but "business improvement" can still encompass using your usage patterns to inform model architecture decisions. Review the full AI Policy for the scope of this carve-out.
Aug 2025 — Harvey + Ironclad Strategic Partnership
Corporate development — Harvey AI + Ironclad announce strategic partnership
+ Harvey and Ironclad announced integration: AI drafting + CLM platform combination flagged: sub-processors, data flow, integration scope impact: Partnership creates a potential new sub-processor relationship. Watch for Ironclad to appear on Harvey's sub-processor list. Data flowing between Harvey and Ironclad would carry both platforms' TOS terms — including Ironclad's §18 $100 liability floor.
Jun 2025 — Harvey + LexisNexis Strategic Alliance
Corporate development — Harvey AI + LexisNexis generative AI + legal content integration
+ Strategic alliance for generative AI integration with LexisNexis legal content flagged: content licensing, sub-processor, LexisNexis data impact: LexisNexis content integration means Harvey's outputs may be derived from LexisNexis-licensed material. Review your LexisNexis license terms — particularly around AI-generated output based on licensed content — before using Harvey + LexisNexis outputs in client-facing work product.
2023 → 2026 — Platform Agreement growth trajectory
harvey.ai/legal/platform-agreement · 4.5x expansion in 3 years · 65+ distinct change events
2023: ~5KB (startup-era boilerplate) + 2024: Expanded to ~12KB — DPA, AI Policy, and sub-processor provisions added + 2025: Expanded to ~18KB — arbitration overhaul, liability floors, LexisNexis/Ironclad partnership provisions + 2026: 23KB+ — EU AI Act, Canadian privacy law, business improvement processing clarification Pattern: Harvey's TOS evolves faster than any other legal tech platform under review. The 15-day sub-processor objection window means firms need a monitoring process — not quarterly reviews.
Get alerted

DriftPatrol watches Harvey's terms so you don't have to.

Get a plain-English Monday brief the day any of these clauses change. Audit trail to the second.

See plans →

Tracked URLs: harvey.ai/legal/platform-agreement · …/data-processing-addendum · …/ai-policy · Informational only — not legal advice.