.png){kind=link}
%201.svg){kind=link}
How Anthropic uses Claude
in legal
Mark Pike, Associate General Counsel, shares how Anthropic's legal team uses Claude to build workflows that cut review times from days to hours—no coding required.
Claude Legal Solutions
Practice better judgment with Claude
Claude handles research, drafting, and the assembly of legal work, so your team can focus on the calls only they can make.
Built for legal work
Claude reviews contracts, surfaces case law, and drafts across the tools your team already uses. Every output can be traced back to its source, on infrastructure your risk and compliance teams trust.
Connected to your tools
Claude connects to iManage, NetDocuments, Docusign, Ironclad, and Thomson Reuters. No copy-paste, context switching, or re-explaining every time you start a new task.
Designed for your practice
New plugins for commercial legal, corporate legal, IP, litigation and more — configured for your practice area, extensible with your team’s playbook and house style. Extend them with custom skills and connectors.
Works where lawyers do
Use Claude for Word, Outlook, and the document management systems your team already uses. In Claude Cowork, delegate multi-step tasks and run skills to speed up your processes.
Confidentiality by design
Enterprise security, full audit trails, no training on your data by default on Team and Enterprise plans. Claude clears the bar for sensitive matters and privileged work, so your team can focus on strategy and high-stakes work, worry-free.
Democratizing access to legal services
We're working with Free Law Project and the Justice Technology Association to make legal help more affordable. Qualifying clinics and public defenders get discounted Claude pricing via the Claude for Nonprofits program.
Claude for the legal industry
New connectors, practice-area plugins, and democratizing access to legal services.
How legal teams use Claude
Research brief — for counsel review
EU AI Act, Article 10
Data minimization obligations for high-risk AI systems
Scope: Art. 10 · DE, FR, IT · 12-month enforcement
Article 10 — key obligations
Providers of high-risk AI systems must meet the following data and governance requirements. Each item is pinned to its source paragraph for verification.
- Datasets must be relevant, representative, and as error-free as possible.
- Verified Reg. (EU) 2024/1689, Art. 10(3)
- Dataset origin, governance, and preparation steps must be documented.
- Verified Reg. (EU) 2024/1689, Art. 10(2)(a)–(c)
- Datasets must be examined for biases affecting health, safety, or fundamental rights.
- Verified Reg. (EU) 2024/1689, Art. 10(2)(f)–(g)
- Special-category data permitted for bias correction only where strictly necessary.
- Confirm Rg. (EU) 2024/1689, Art. 10(5); cross-ref. GDPR Art. 9(2)(g)
- Operates alongside GDPR Art. 5(1)(c) — does not displace minimization.
- Verified AI Act Recital 67; cross-ref. Reg. (EU) 2016/679, Art. 5(1)(c)
Art. 10 obligations cascade into Art. 9 (risk management) and Art. 17 (quality management). Defined terms "training data," "validation data," and "testing data" (Art. 3) carry through; verify consistency where source materials predate the final adopted text.
Member state implementation
Germany
- Notable posture BNetzA leads market surveillance; emphasis on dataset documentation.
- Open issue BNetzA / Länder DPA allocation unsettled.
- Source pin BMWK guidance, Mar. 2025
France
- Notable posture CNIL active; strict-necessity test for Art. 10(5) processing.
- Open issue National AI authority designation pending.
- Source pin CNIL recommendation, 2024
Italy
- Notable posture Garante enforcing on training-data lawfulness ahead of phase-in.
- Open issue Sequencing with AI Act timelines unclear.
- Source pin Garante decision register
Cross-references not fully traced: French CNIL recommendations predate the final AI Act text — check whether "high-risk" scope mapping still aligns. Italian Garante decisions cite GDPR, not the AI Act; treat as analogous, not directly on point.
Recent enforcement — last 12 months
- Garante (Italy). Training-data lawfulness in generative AI deployment. Penalty amount — confirm before relying.
- Verified Garante decision, 2024 (registry no. — confirm)
- CNIL (France). Sanction on dataset documentation and lawful basis. Penalty amount — confirm before relying.
- Confirm CNIL deliberation — reference not located
- EDPB. Opinion on AI training data and GDPR minimization — guidance instrument.
- Verified EPB Opinion 28/2024 (confirm number)
Flag for counsel — unsettled and cross-cutting
Items below are unresolved at the regulatory level or depend on cross-references that should be re-traced before reliance.
- Scope of Art. 10(5) strict-necessity test — guidance varies; cross-check against GDPR Art. 9(2)(g) and member state implementing law.
- Interplay with GDPR Art. 5(1)(c) — not yet litigated; AI Act Recital 67 signals concurrent application but does not resolve overlap.
- Documentation depth expected under Art. 10(2) — divergent national posture likely; BNetzA, CNIL, and the Garante have not aligned.
- Defined-term consistency — "training data," "validation data," "testing data" (Art. 3) may not map cleanly onto pre-existing dataset taxonomies in client materials.
Legal review required. Claude-generated content — verify conclusions and citations before relying.
Legal research
Ask Claude to surface relevant case law or regulatory guidance via Thomson Reuters. Get a structured memo with citations, not a list of links to dig through yourself.
Due diligence document review
Connect Claude to a Box or Intralinks data room. Claude reads and categorizes the document set, flags material issues, and drafts the diligence summary. Work that used to take days of associate time.
Statement of Work
SOW No. 2026-DV-014 · Issued under MSA dated March 14, 2024
1. Term and scope
Six-month engagement under the MSA. Vendor shall provide data labeling services for ten thousand (10,000) documents per calendar month, with completed datasets delivered to Client no later than the 15th of the following month per Exhibit A.
2. Fees and payment
Fees shall be invoiced monthly in arrears upon acceptance. Payment terms shall be Net 601 from date of invoice.
- 1 Counterparty changed Net 30 to Net 60. Last 8 vendor SOWs all closed at Net 30 with no concession.
- Pinned to template §5.2
3. Intellectual property
All labeled datasets and derivative outputs produced under this SOW shall be the sole property of Client. Vendor retains a perpetual, royalty-free license to use anonymized annotation methodology and aggregated statistical outputs for internal model improvement.2
- 2 Counterparty added IP license-back. Template assigns all work product to Client outright; this carves out methodology and aggregated outputs.
- Cascades to §9 confidentiality — "Aggregated Outputs" undefined
4. Indemnification
Vendor shall indemnify Client against third-party claims arising from Vendor's gross negligence or willful misconduct3.
- 3 Indemnification narrowed from broad performance/IP/confidentiality to gross negligence and willful misconduct only.
- §11.3 cross-reference may break without conforming edit
5. Governing law
This SOW shall be governed by the laws of the State of California4, without regard to conflict of laws principles.
- 4 Governing law changed from Delaware to California. We've accepted CA on 2 of 12 prior SOWs when counterparty held leverage.
- MSA §22 venue still references Delaware courts
Legal review required. Claude-generated content — verify conclusions and citations before relying.
Contract lifecycle
Claude drafts the NDA or SOW in Word, routes it to Ironclad for CLM tracking, and surfaces status on request. The whole flow without leaving Claude.
Outside counsel and matter review
Pull billing and matter data into Claude. Spot off-pattern entries, flag matters drifting against budget, and prep for relationship-partner conversations.
Regulatory compliance gap analysis
Run a compliance gap analysis against a named framework. Claude pulls evidence from your systems, scores severity, and flags what needs attention before audit.
Build legal products with Claude
Embed Claude into your legal platform or product through the API and Agent SDK.
MCP for legal data
Connect your contract systems, document repositories, and matter data to Claude through Anthropic's open Model Context Protocol.
Frontier models
Claude leads on legal reasoning benchmarks across contract review, citation accuracy, and long-document comprehension.
Built for sensitive data
Deploy in sandboxed containers with network controls, credential isolation, and audit logging built into the SDK.
Tools made for the way you work
Claude for M365
Word is where legal work gets done. Claude handles full redlining with tracked changes, document comparison, and MSA-scale review without leaving the doc. It pulls in pleadings, prior versions, and Outlook context, drafts the next turn for your review, and tracks every change for the audit trail.
Claude Cowork
Hand off multi-document matter tasks: drafting a memo against pleadings, preparing for a deposition. Like a trusted senior associate, you delegate to Claude and come back when it’s done, so you can do deeper work.
Plugins
Add practice-area plugins for contract review, NDA triage, compliance workflows, and more, all configurable to your organization's playbook and risk tolerances.
Platform
Integrate Claude directly into your CLM, matter management, or e-discovery systems. Build production-grade workflows on the same frontier models that firms like Freshfields and Quinn Emanuel do.
Legal resources
How Anthropic's legal team cut review times from days to hours with Claude
Blog
How Claude helps legal teams handle the work that needs your judgment
Blog
Watch the Claude for Legal teams webinar
Webinar
Prep scattered documents for a compliance audit
Use case
Organize your legal workflows using Projects
Use case
Product guide: Claude for the legal industry
Guide
See what Claude does for legal teams
Whether you're an in-house team looking to scale or a firm building on Claude, we can help you figure out where to start.