Most lawyers who adopt AI tools spend significant time evaluating features and pricing. Most spend almost no time evaluating security. This is backwards.
Legal information is among the most sensitive data categories in any industry. Client communications, M&A strategy, litigation positions, personal injury records — these are documents whose unauthorized disclosure can end careers, tank deals, and expose firms to malpractice liability.
This guide provides a systematic security evaluation framework for legal AI tools, built around 8 questions that any vendor should be able to answer.
Why Security Evaluation Is Different for Legal AI
Professional responsibility obligations. Lawyers in most jurisdictions have a duty of competence that extends to understanding the technology they use, and a duty of confidentiality that requires taking reasonable steps to protect client information.
Attorney-client privilege. Uploading privileged communications to a third-party AI platform raises questions about whether the privilege is waived. See our attorney-client privilege AI guide for the legal framework.
AI training data. Several major AI vendors have updated their terms to expand their rights to use customer inputs for model training. Some legal AI vendors have explicit "no training on client data" commitments. Others do not.
Cross-border data flows. Where your AI vendor processes and stores data matters for GDPR compliance and local data residency requirements.
The 8-Question Security Evaluation Framework
Question 1: Is the vendor SOC 2 Type II certified?
What it means: SOC 2 Type II is an independent audit verifying that a vendor's security controls are in place and operating effectively over time.
What to look for: Request the SOC 2 Type II report, not just a statement that they are certified.
Red flag: Vendors who say they are "working toward" certification haven't had their controls independently verified.
Tools with documented SOC 2 Type II: Harvey AI, Luminance, ContractPodAi, Relativity, Clio.
Question 2: Does the vendor use client data to train AI models?
What it means: Some vendors use customer interactions to improve their models — your client documents may become part of training data.
What to look for: A specific contractual commitment in the Terms of Service or DPA: "We do not use Customer Data to train, fine-tune, or improve our AI models."
Tools with public no-training commitments: Harvey AI, Clio, Relativity, Paxton AI.
Question 3: What is the vendor's data retention policy?
What it means: How long does the vendor store your queries, documents, and AI outputs? Zero data retention means the vendor deletes everything after each session.
What to look for: Ask specifically: how long is my data retained? Where is it stored? Can I request deletion?
Question 4: Will the vendor sign a Data Processing Agreement?
What it means: A DPA defines how the vendor handles your client data, what security measures they maintain, and their obligations in a breach event.
What to look for: Any vendor processing EU personal data on your behalf must provide a DPA under GDPR Article 28. See our Data Processing Agreement glossary entry.
Question 5: What are the vendor's breach notification obligations?
What it means: If the vendor suffers a breach that exposes your client data, how quickly must they notify you?
What to look for: A contractual commitment to notify you within a defined timeframe (typically 72 hours under GDPR) with specific information about what data was affected.
Question 6: Where is data processed and stored?
What it means: In which jurisdictions do the vendor's servers operate?
What to look for: Specific disclosure of data center locations. For EU matters, look for EU-specific data residency options or Standard Contractual Clauses. See our Standard Contractual Clauses entry.
Question 7: What access controls and audit logging are in place?
What it means: Who within the vendor organization can access your firm's data?
What to look for: Documentation of internal access controls. See our audit log entry for what to look for.
Question 8: Is there an on-premise or private deployment option?
What it means: Some AI vendors offer deployment on your firm's own infrastructure — your client data never leaves your environment.
What to look for: Documented on-premise deployment options. See our on-premise AI entry. Tools offering this: Harvey AI, Luminance.
The Security Evaluation Process: Step by Step
Step 1: Tier your matters by sensitivity. Not all client data requires the same security level.
Step 2: Send a vendor security questionnaire. Ask all 8 questions above. Document the responses in writing.
Step 3: Review the Terms of Service and DPA. Look specifically for training data provisions, breach notification obligations, and data retention terms.
Step 4: Verify certifications. SOC 2 Type II and ISO 27001 are verifiable. Ask for the report date and scope.
Step 5: Establish an internal AI use policy. Document which tools are approved for which matter types. See our Law Firm AI Policy entry for the framework.
FAQ
Q: Does uploading documents to an AI tool waive attorney-client privilege?
It depends on the vendor agreement and jurisdiction. Well-structured vendor agreements treat the AI tool as a confidential agent of the law firm, preserving privilege. Verify the privilege treatment in any vendor's Terms of Service before uploading privileged materials.
Q: Is GDPR compliance required for US law firms?
US law firms that process personal data of EU clients or individuals are subject to GDPR, regardless of where the firm is located. See our GDPR compliance AI tools entry.
Q: How do I handle AI security for healthcare law matters?
Healthcare legal matters involving PHI trigger HIPAA obligations. You need a Business Associate Agreement with any AI vendor processing PHI. See our HIPAA AI legal tools entry.
Q: What if a vendor fails to meet my security requirements?
Use a more secure alternative for that matter type. Declining to use a tool for matters where it doesn't meet your security standards is the correct professional response.
Q: How often should I re-evaluate AI vendor security?
At minimum annually, and whenever a vendor updates their Terms of Service, experiences a publicly reported security incident, or significantly changes their AI architecture.
This guide reflects editorial research and does not constitute legal advice. Consult qualified legal counsel for jurisdiction-specific professional responsibility guidance. Scores reflect LawyerAI editorial assessment as of May 2026.