The most common Rule 1.6 violation in AI use isn't deliberate. It's an attorney who uploads a client contract to a general-purpose AI tool to get a summary, without realizing that the tool's default settings include training on user inputs.
That input — which may contain the client's name, the counterparty, deal terms, financial information, and strategic context — is now potentially part of a training dataset that could surface in another user's output. The attorney didn't intend to disclose client information. The attorney also didn't ask.
What AI Platforms Actually Do with Your Data
Different AI tools have very different data handling practices, and those practices are not always prominently disclosed.
Consumer-facing AI tools — the versions of ChatGPT, Gemini, and similar tools available through standard free or low-cost accounts — typically reserve the right to use conversation content for model training. Enterprise versions of these same tools typically include contractual commitments that prohibit training on user data. The difference between the two is often nothing more than which URL the attorney navigated to.
Beyond training data practices, there are retention questions: How long is the conversation stored? Who at the vendor organization has access to it? What happens to the data if the attorney terminates the subscription? What are the vendor's breach notification obligations?
ABA Formal Opinion 512 is explicit that attorneys must understand the data handling practices of any AI tool before using it with client information. "I didn't know" is not a defense under Rule 1.6 any more than it would be for leaving client files in a public location.
What a Vendor Review Should Cover
Before approving any AI tool for use with client information, a governing firm should document answers to the following questions:
Data retention: How long does the platform retain conversation inputs? Is there a configurable retention period? What triggers deletion?
Training data: Does the platform use customer inputs for model training? Is this opt-out or opt-in? What data is excluded from training?
Access controls: Who within the vendor organization can access stored conversations? Under what circumstances? What logging exists?
Breach notification: What are the vendor's contractual obligations if the platform experiences a breach involving customer data? What is the notification timeline?
Contractual protections: Does the vendor offer a data processing agreement (DPA) or similar contractual instrument? Does it include confidentiality provisions adequate for legal client data?
Subprocessors: Does the platform route data through third-party AI models or infrastructure? What are those subprocessors' data handling commitments?
This is not an exhaustive list. It is the minimum due diligence threshold before using any AI platform with identifiable client information.
The Informed Consent Question
Even with appropriate vendor controls in place, there may be circumstances where client informed consent is appropriate before using AI tools in connection with a matter. This is particularly true for highly sensitive matters — criminal defense, family law, executive employment disputes — where the client's interest in the confidentiality of their information is especially acute.
Some state bars have begun issuing guidance on AI use and client disclosure. The trajectory is clear: the profession is moving toward a norm in which clients are informed, in reasonable terms, about how AI tools are used in their matters and what protections are in place.
Firms that build this disclosure into their engagement agreements now are ahead of what will eventually be a required practice.
What Governance Looks Like
A Rule 1.6-compliant AI governance posture has two layers: pre-approval of tools, and per-matter data handling protocols.
At the tool level, every AI platform used by the firm should be reviewed against the criteria above before any client data is introduced. Approved tools should be documented in writing. Unapproved tools should be explicitly prohibited for use with client information.
At the matter level, the firm should have a defined protocol for what types of client information can be used with AI tools, under what conditions, and with what verification steps. This protocol should be part of the firm's supervision framework and reflected in training for attorneys and staff.
The AI Governance Phase 0 Assessment measures both layers under the Confidentiality Controls and Tool Approval & Access scoring domains. A score below 2.0 in either domain indicates that the firm's current posture does not meet the minimum threshold for defensible AI use under Rule 1.6.