You have been using generative AI to do your job better. You asked it to turn a 300-line bug spreadsheet into a readable executive summary for your leadership team. You used it to draft test plans for a new procurement portal. You pasted in your quarterly audit findings and asked for a remediation priority matrix. You may have even used it to draft your organization’s multi-year accessibility roadmap.
None of that felt risky at the time. You were busy, and it felt efficient.
Then the demand letter arrived.
In retrospect, you may view the wisdom of using an AI tool differently.
This article is written specifically for accessibility managers, coordinators, and program leads who are not attorneys but whose employers may receive demand letters alleging ADA or Section 508 violations and have been using public generative AI tools in their accessibility work. The legal concepts here overlap with attorney-client privilege, but the more immediate problem for you is different: you may have created a paper trail of your organization’s accessibility failures and handed it to a third-party commercial platform before you ever talked to a lawyer.
As always, none of the information contained in any of my blogs should be viewed as legal advice.
What a Demand Letter Actually Means
A demand letter is not a lawsuit; it is a threat of one. It is a formal written notice, usually from an attorney representing a person with a disability or a disability rights organization, asserting that your organization has violated the ADA, Section 508, or a state accessibility law and demanding a specific remedy. Commonly demanded remedies include a compliance timeline, an accessibility audit, remediation of specific violations, and monetary compensation.
Your organization is not required to respond immediately, or at all. You absolutely should not respond without involving your organization’s legal counsel. If your organization does not have in-house counsel, the demand letter must be sent to outside counsel experienced in accessibility law before you take any action.
What makes demand letters legally significant from an evidence standpoint is that they mark a point in time at which your organization had documented, specific notice of alleged accessibility barriers. Everything your organization does from the moment that letter arrives is potentially discoverable if litigation is filed, and also relevant to whether your organization acted in good faith.
What is Attorney-Client Privilege
Attorney-Client Privilege (sometimes called ACP) is the evidentiary rule that protects confidential communications between an attorney and a client made for the purpose of obtaining or rendering legal advice. The privilege belongs to the client and prevents those communications from being compelled as evidence in litigation. ACP only exists when the communication is confidential
Your GenAI Prompts are Evidence
When you fed a list of 300 accessibility bugs into ChatGPT and asked it to produce an executive summary, you did something with real legal consequences. You created a readable, organized, concise document summarizing the accessibility failures your organization was aware of. That document and the underlying bug list are evidence of your organization’s actual knowledge of the barriers underlying the demand.
In ADA litigation, the question of what an organization knew and when they knew it is central to liability analysis, particularly for claims involving deliberate indifference or pattern-and-practice violations. A well-organized executive summary of known accessibility bugs, generated by AI and distributed to leadership, and then not acted on, is potent evidence that the organization was aware of widespread barriers and had documented them.
That summary did not remain within your organization. It was generated by transmitting your bug list to a third-party commercial platform and was processed by a system your organization does not own or control.
Test plans generated with AI assistance are another category of potential evidence. A well-constructed test plan describes what was tested, which criteria were applied, which components of the system were in scope, and, implicitly, what was not tested or not yet remediated. For example, a test plan you generated for a procurement portal documents that you were evaluating the portal for accessibility compliance at a particular point in time.
If the demand letter alleges barriers in that portal, the test plan becomes relevant to questions such as: when the organization assessed the portal for accessibility, what it found, and what it did about it. Also, did you follow the AI’s recommendation, or did you shorten the test plan to meet a compressed delivery schedule? A test plan stored on a shared drive is already a discoverable document within your organization’s control. The additional problem introduced by AI is that the chain of thought used to develop the test plan is now discoverable on an external, non-confidential platform.
Your Roadmaps and Priority Matrices Reveal Strategic Decisions
Remediation roadmaps and priority matrices are especially sensitive in litigation because they reveal how your organization prioritized what to fix first. If your AI-assisted priority matrix placed certain barriers in a low-priority category and those barriers are the ones described in the demand letter, you have a document showing that your organization consciously chose to delay addressing those specific issues. Now the onus is on your organization to argue that those low-priority issues didn’t block the user.
Courts and opposing counsel understand remediation planning documents. They know that when an organization creates a multi-year roadmap for accessibility compliance, it has accepted the risk of operating with known barriers in the interim. That is a legal exposure, and your AI-generated planning documents may make it easy to demonstrate.
The Attorney-Client Privilege Problem After You Receive a Demand Letter
Now that you have received the demand letter and retained counsel, your organization is in a very different legal posture. Communications between you and your attorney, made in confidence to obtain legal advice, are protected by the attorney-client privilege. Your attorney’s mental impressions, legal theories, and strategic analysis may be protected as work product.
But here is the critical point many accessibility managers do not understand: if you continue to use public generative AI tools in connection with your response to the demand letter, you may waive the privilege that protects those communications.
One incredibly common request is for the attorney to ask you to compile a list of all known accessibility barriers in your customer-facing web properties. You have a spreadsheet. You use ChatGPT to organize it and draft a summary before sending it to your attorney. That summary, prepared to facilitate legal advice, would normally be protected by the attorney-client privilege. But by running it through a public AI tool first, you transmitted it to a third party outside the attorney-client relationship. The privilege analysis of that document has become complicated at best and compromised at worst.
The same concern applies to drafting correspondence with opposing counsel, preparing documentation in response to a complaint, developing your organization’s public-facing accessibility statement as part of a litigation response, and creating any internal documentation that your attorney relies on to advise you.
From the moment you engage counsel in connection with a demand letter, treat every document, email, or conversation in connection with that matter as potentially privileged. Stop running queries through a public AI tool. This doesn’t mean you must stop using AI entirely. For tasks that are genuinely disconnected from the matter at hand, including drafting training materials on unrelated topics, writing general accessibility guidance for your team, writing reviews for team members, or conducting research on accessibility best practices as a general matter, the privilege concerns do not apply in the same way. But anything that touches the subject matter of the demand letter needs to stay out of public AI tools.
Litigation Holds and AI-Generated Documents
As one of the first steps in responding to a demand letter, your attorney should advise you to implement a litigation hold. A litigation hold is a directive to preserve all documents and data that may be relevant to the dispute, suspending your organization’s normal document retention and deletion policies.
AI-generated queries and documents are not exempt from litigation holds. If you have executive summaries, test plans, roadmaps, or any other accessibility-related documents created with AI assistance, you must preserve them once you become aware of potential litigation. This includes the original prompts you used to generate them.
Another thing many accessibility professionals miss is that it’s not just ChatGPT and Claude. Some organizations use AI tools to generate notes, action items, and summaries from online meetings and store them in users’ accounts. If you have used such a tool and your conversation history is accessible, your attorney needs to know. Those logs may be discoverable, though obtaining them from a third-party AI vendor involves a range of procedural and contractual issues your attorney will need to navigate.
Do not delete anything once you receive a demand letter or notice of litigation. Do not clear your AI conversation history. Do not reformat or revise AI-generated documents that may be relevant to the dispute. Spoliation, the legal term used to describe destroying relevant evidence, carries serious consequences, including adverse inference instructions that permit a court to tell a jury that the destroyed evidence would have been unfavorable to your organization.
The Deeper Lesson: AI Use Requires a Legal Framework
For many accessibility managers, the hardest part of this situation is that nothing they did felt wrong at the time. Using AI to find patterns in a 300-line bug spreadsheet is not misconduct. Writing AI-assisted test plans is a legitimate productivity strategy. Building an accessibility roadmap with AI assistance is reasonable program management.
The problem is that the legal implications of those choices were not apparent until the demand letter arrived. Most accessibility managers are not lawyers and do not approach their work through the lens of evidence, privilege, and discovery. They focus on WCAG conformance levels, assistive technology compatibility, and remediation timelines. Legal risk is someone else’s department.
But accessibility law has matured significantly. The DOJ’s Title II rule, the steady growth of ADA Title III litigation, and the increasing sophistication of plaintiffs’ attorneys mean that accessibility managers at organizations of any meaningful size should expect to encounter formal legal processes at some point. A demand letter is not a sign that your organization is terrible at accessibility. It is an increasingly routine part of the legal landscape.
The lesson is not to stop using AI. It is to build a legal framework for how AI is used in accessibility work before a demand letter arrives. That means understanding what information you share with AI platforms, establishing an AI use policy that addresses sensitive program data, confirming whether your AI tools have enterprise agreements with confidentiality protections, and maintaining a relationship with legal counsel who understands both accessibility law and technology. Organizations that proactively build that framework will be in a much better position when a demand letter arrives than those that try to figure it out after the fact.
Receiving demand letters is an all-too-common part of accessibility managers’ jobs. Organizations resolve these matters every day through negotiated compliance agreements, corrective action plans, and demonstrated good-faith remediation. However, your ability to demonstrate good faith and to protect the strategic advice your attorney provides depends in part on understanding what you have already entered into the AI and ensuring you do not compound the exposure going forward.
You are not a lawyer. You do not need to become one. But you do need to understand that the tools you use to do your job create a record, and that record matters when the law gets involved.