May/June 2026 Issue

Also in this issue: Municipal Regulation of Data Centers     |     Public Bidding Refresher: Changes to Public Bidding Threshold and Best Practices Reminders

AI and Attorney-Client Privilege: What Wisconsin Municipalities Need to Know

Emmerson Mirus | 06.01.26

Municipalities are introducing artificial intelligence (AI) tools in their everyday operations — from drafting reports and correspondence to processing citizen inquiries to optimizing waste collection or public transit routes. While AI tools offer efficiency gains, they can also introduce significant and underappreciated legal risks. Chief among these risks is the potential loss of attorney-client privilege and work product protection. For Wisconsin municipalities, transparency obligations intersect with litigation exposure. Therefore, improper or careless AI use may transform confidential legal strategy into discoverable — or even public — records. Understanding these risks is critical to preserving attorney-client privilege and institutional stability. 

Attorney-client privilege protects confidential communications between a client and an attorney made for the purpose of obtaining legal advice from disclosure. As such, the privilege exists where communications are (1) between a client and a lawyer or their representatives, (2) intended to be confidential, and (3) made to secure legal advice.

Generative AI tools complicate these protective frameworks because users’ interactions with them may fall outside the traditional scope of privilege. A recent federal case, United States v. Heppner¹, illustrates the risk. There, a criminal defendant used Anthropic’s Claude to generate documents analyzing his legal exposure and defense strategy. The court held that the attorney-client privilege did not protect those documents. Applying the first element of attorney-client privilege, the court explained that the AI tool was not a lawyer, so the communications were not between the defendant and his attorney. Likewise, because the AI tool’s privacy policy explicitly stated that it collects, stores, and trains its model on user inputs and outputs, there was no reasonable expectation of confidentiality. Finally, the court emphasized that the AI platform itself disclaimed any ability to provide legal advice.

The Heppner case underscores that sharing information with an AI tool may be treated as a voluntary disclosure to a third party, which in turn can waive privilege altogether. Courts commonly hold that privilege is lost when confidential information is shared with those outside the attorney-client relationship. Because AI tools — particularly consumer-grade ones like ChatGPT or Gemini — are generally not bound by duties of confidentiality and may store, process, or disclose user inputs, legal strategy, factual summaries, or internal communications submitted to them may become discoverable in litigation.

Inadvertently surrendering privilege can be devastating in litigation. Municipalities routinely face claims involving disputes with public employees, civil rights allegations, land use decisions, and contract disputes. If internal legal analyses or litigation strategies are disclosed through AI use, opposing parties may gain direct access to the municipality’s thought process and legal positioning. In practical terms, this can weaken defenses, undermine settlement leverage, and increase exposure to liability.

More specific to Wisconsin municipalities, AI-generated content may itself be subject to disclosure under Wisconsin’s Public Records Law. Wisconsin defines ​“record” broadly to include “[a]ny material on which written, drawn, printed, spoken, visual or electromagnetic information is recorded or preserved, regardless of physical form or characteristics, which has been created or is being kept by a [ ] [governmental] authority.”² Because public officials’ prompts, outputs, and related communications with AI tools are ​“created” and may be stored or retrieved, they could be subject to disclosure upon request unless a specific exemption applies. The Public Records Law presumes openness and requires authorities to justify withholding records through statutory exceptions or a balancing test. If privilege has already been waived due to improper AI use, municipalities may have little basis to withhold records created with or by AI from public disclosure.

Generative AI presents both opportunity and risk for municipalities. While AI does not need to be avoided altogether, its use must be deliberate and controlled. Heppner demonstrates that courts have started applying traditional privilege principles to these new technologies. As such, municipal officials and counsel should treat public AI tools as third parties for privilege purposes and avoid inputting confidential or legally sensitive information absent appropriate safeguards. Municipalities that fail to adapt may see their most sensitive legal communications transform into trial evidence — or worse, press releases.