Shareholder
Share this article:
AI Tools in CIDs: User Beware
For better or worse, artificial intelligence has arrived in our industry. Both board members and homeowners alike across California are turning to large language models (LLMs) like ChatGPT and Claude to navigate the complicated world of community association governance. While convenient for some, the risks are real. If used carelessly, AI can expose confidential information, generate legally inaccurate advice and information, and arm adversarial homeowners with enough inaccurate (and lengthy) ammunition to keep a board busy and asking for legal intervention and interpretation. In this article we highlight the pitfalls of AI and provide tips on how to avoid them.
Preserving Executive Session Confidentiality and Attorney-Client Privilege
Boards members must exercise serious cautious when using AI. The California Civil Code allows boards to hold executive sessions, also known as closed session, meetings to discuss specific sensitive matters like litigation, the formation of contracts, member discipline, personnel matters, and delinquencies. These sessions exist precisely because the California legislature recognized that the topics should remain confidential. What a board discusses in executive session cannot be disclosed to the general membership without board authority.
In light of that, when a board member copies and pastes executive session notes or details of pending litigation, for example, into a commercial AI tool to get a summary or draft a response, they may be unwittingly pushing that information into a system they do not control. Most consumer-facing LLMs retain conversation data to some degree and use inputs for training unless users opt out. The LLMs store information on servers governed by terms of service, not California law. As such, using LLMs for association purposes may unwittingly breach the confidentiality of executive session without board authorization.
This exposure is not hypothetical. Indeed, in the federal case of United States v. Heppner (2026), the court ruled that documents generated using a publicly available AI tool are not shielded by attorney-client privilege or the work product doctrine. In the case, Heppner used an AI tool for guidance on his legal case. Heppner created documents through the AI tools and shared them with his legal counsel. The court ruled that the documents created were not protected by attorney-client privilege. While the court limited the ruling, acknowledging that the outcome may differ with alternative facts – for example, if the attorney had used the AI tool – and this case is not direct precedence in California, this case is a warning for all boards. If you share what you consider confidential information with an LLM, there is little protection for the work product that results. Exposing confidential information and waiving attorney-client privilege without board authority could be seen as a breach of fiduciary duty to the association.
PRACTICE TIP: Never input names, addresses, account balances, legal strategy, personnel matters, or any information discussed in executive session into a commercial AI tool. Use AI for structure and language – not processing content or asking for specific legal advice. |
When Owners Use AI Against the Board
We have recently seen in our practice an uptick in homeowners using AI tools strategically against the association. Homeowners who believe the board has violated their rights are armed with open-source summaries that have boards working overtime. Being informed about the applicable law is not inherently problematic. Informed homeowners will hopefully make for healthier communities. However, AI-generated content is not always accurate. This is because many AI tools are designed in a sycophantic nature; meaning these tools are designed to generate responses the system thinks the prompter wants, as to opposed to providing the most accurate and correct response. The results produced are produced with such clarity that homeowners have been known to use legal misinformation with confidence and stubbornness. When boards attempt to correct the record or provide the necessary context, the perception of bad faith can spiral quickly.
In this upswing in AI usage, we have also received lengthy association record demands as the AI tools dramatically lower the effort required to generate formal-sounding demand letters and Public Records Act-style document requests. A single motivated homeowner can now produce a volume of written demands that would previously have required legal representation. Management companies are flooded with correspondence that is time-consuming to answer, even when meritless.
Conclusion
While it is unlikely that boards or owners will abandon AI tools, both sides should use them with clear-eyed awareness of the limitations. Boards may want to consider adopting an AI policy determining which tools are approved and how. Owners should treat AI output as a starting point for research, not a final legal opinion. For both owners and boards, when a dispute may have serious consequences, reach out to legal counsel. It is worth the cost to double check the advice from AI and to be able to rely on attorney-client privilege to protect the advice given to the board.


