AI Governance & Quality Control
Protecting Client Confidentiality When Law Firms Use Cloud AI
By Anthony Villanueva ·
Cloud-based AI tools process client data on third-party infrastructure. ABA Model Rule 1.6 and state bar guidance require attorneys to take reasonable measures to protect that data. This guide defines what 'reasonable measures' means in the AI context.
ABA Model Rule 1.6 requires attorneys to take reasonable measures to prevent unauthorized disclosure of client information. For cloud-based AI tools, this means evaluating how each tool processes, stores, and uses client data before permitting its use in client matters — not after. The professional obligation is proactive, not reactive: attorneys must assess confidentiality risk before using a tool, not discover it after a breach.
Core questions for confidentiality analysis
The confidentiality analysis for a cloud AI tool should address four questions. First, is client data used to train the tool's underlying model? If a tool uses client data to fine-tune its model and makes that model available to other users, client confidential information could be indirectly exposed through the model's outputs for third parties. Most enterprise legal AI platforms explicitly exclude client data from model training, but this should be verified in the vendor's data processing agreement, not assumed.
Second, where is client data stored, and what are the data residency requirements for the relevant jurisdictions? Third, what access controls govern who can access client data within the vendor's infrastructure? Fourth, what breach notification obligations apply if client data is compromised?
Vendor contract requirements for data protection
The vendor contract should include explicit representations on each of these questions, along with specific provisions that address legal-industry-specific requirements: prohibition on model training with client data, data deletion obligations at contract termination, breach notification timelines consistent with applicable bar requirements, and indemnification provisions for confidentiality breaches caused by the vendor's system.
Applying NIST frameworks to third-party risk
NIST's AI Risk Management Framework's guidance on third-party AI risk is a useful supplement to the professional responsibility framework for this analysis. NIST treats third-party AI vendors as introducing risk into an organization's AI portfolio that must be assessed, managed, and monitored — not just approved once and forgotten.
The ongoing monitoring obligation is often overlooked in law firm AI governance: vendor security certifications (SOC 2, ISO 27001) expire and must be renewed; data handling practices can change with product updates; and a vendor's security posture at time of approval may not reflect its current state.
Frequently asked questions
What does ABA Model Rule 1.6 require regarding cloud-based AI tools?
ABA Model Rule 1.6 requires attorneys to take reasonable measures to prevent unauthorized disclosure of client information. For cloud-based AI tools, this means evaluating how each tool processes, stores, and uses client data before permitting its use in client matters — not after. The professional obligation is proactive, not reactive: attorneys must assess confidentiality risk before using a tool, not discover it after a breach.
What are the four questions that must be addressed during the confidentiality analysis for a cloud AI tool?
The confidentiality analysis for a cloud AI tool should address four questions. First, is client data used to train the tool's underlying model? If a tool uses client data to fine-tune its model and makes that model available to other users, client confidential information could be indirectly exposed through the model's outputs for third parties. Most enterprise legal AI platforms explicitly exclude client data from model training, but this should be verified in the vendor's data processing agreement, not assumed.
Second, where is client data stored, and what are the data residency requirements for the relevant jurisdictions? Third, what access controls govern who can access client data within the vendor's infrastructure? Fourth, what breach notification obligations apply if client data is compromised?
What is an often-overlooked obligation in law firm AI governance regarding third-party vendors?
NIST's AI Risk Management Framework's guidance on third-party AI risk is a useful supplement to the professional responsibility framework for this analysis. NIST treats third-party AI vendors as introducing risk into an organization's AI portfolio that must be assessed, managed, and monitored — not just approved once and forgotten.
The ongoing monitoring obligation is often overlooked in law firm AI governance: vendor security certifications (SOC 2, ISO 27001) expire and must be renewed; data handling practices can change with product updates; and a vendor's security posture at time of approval may not reflect its current state.