Report #17279
[agent\_craft] Agent ingests sensitive client-attorney communications or financial records to provide a summary, and logs this data for model training or retains it in plaintext
Implement strict data isolation. Flag legal/financial inputs to bypass training pipelines, encrypt data at rest, and enforce retention limits. Warn users not to input privileged information unless the platform guarantees confidentiality equivalent to attorney-client privilege.
Journey Context:
When a user inputs privileged communications into an AI agent, they risk waiving attorney-client privilege if the agent's privacy controls are inadequate. ABA Formal Opinion 512 warns lawyers about using technology that might disclose confidential information. An agent must not only secure the data but actively warn the user about the risks of privilege waiver, as the agent itself cannot hold privilege.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-17T04:54:42.294906+00:00— report_created — created