Business and Legal Insights

Artificial intelligence has become a part of everyday business. Employees regularly use AI to summarize documents, draft e-mails, analyze contracts, and even attempt to understand legal issues. While AI tools offer significant benefits, they also create legal risks that many businesses overlook. Before using legal, confidential or customer information with any AI system, users should understand what they are agreeing to and, when in doubt, consult their attorney first.

One of the biggest mistakes a law firm client can make is copying legal advice or work product from its attorney into a public or consumer AI tool such as ChatGPT, Claude, or Gemini.  Doing so not only poses a risk to the confidentiality of whatever information is disclosed, but also puts the attorney-client privilege at risk.  Courts are only now beginning to address these questions, so clients cannot afford to wait for the law to catch up before taking precautions, and educating employees about these issues.

In February 2026 a federal court in New York held that information submitted by a user to a public AI platform on their own accord, before engaging a lawyer, was not protected by either the attorney-client privilege or the work-product doctrine. Other courts, by contrast, have declined to find that using a public AI tool automatically waives the privilege or work-product protection. Courts are also starting to explore whether entering trade secrets into a public AI system means that the information, invention or data is no longer a trade secret. Because the law will remain unsettled for a few years, it is important to note that submitting an attorney’s analysis to a public AI tool may risk waiving the attorney-client privilege and may lead to loss of confidentiality or trade secret protection.  Given that uncertainty, it is prudent to be cautious, rather than take risks with confidential company information.

Loss of Attorney-Client Privilege

Communications between a client and its attorney are generally protected by the attorney-client privilege, but only when those communications are “not intended to be disclosed to third persons” other than those reasonably necessary to render legal services. Sharing privileged communications with a public AI tool operated by a third party may fall outside this protected circle.  Determining whether use of a particular AI tool waives the privilege may depend on the AI tool’s terms of service, privacy practices, and how the information is processed or retained.

It is likely that courts will find a distinction between public, consumer-grade AI tools and enterprise or “closed-box” versions. Public AI platforms typically retain user inputs, may use them for model training, and reserve the right to disclose data to third parties, all of which undermine a claim to confidentiality. By contrast, enterprise versions of tools such as ChatGPT Enterprise, Claude for Business, and other closed-box AI platforms may contractually prohibit the vendor from retaining or using inputs for training, and may restrict third-party disclosures. There is still a risk that the data may be accessed without a business’s knowledge or consent, including by hackers or rogue employees, but whether a particular enterprise AI tool preserves confidentiality is a fact-specific inquiry that depends on the contractual terms in place.  It’s also unclear how reliable each AI tool’s safeguards are, as evidenced by several examples where a closed tool ended up sharing information outside of the platform.

The work-product protection – which shields an attorney’s litigation strategy and mental impressions – follows a different standard.  Some courts have held that it is waived only by disclosure to an adversary, and that using an AI tool does not necessarily waive such protection. However, these doctrines will also continue to develop in the coming years.

Loss of Confidentiality or Trade Secret Protection

Even if attorney-client privilege is not at issue, business employees should still beware that inputting confidential information, such as draft agreements, business strategies, financial information, customer data, trade secrets, and litigation strategy, into any AI tool creates risk. Many AI tools process information on servers outside of your organization. Some of these providers may retain prompts for a period of time.  Others may use submitted information to improve their models unless those features are disabled or an enterprise version is used.

Under federal and state trade secret laws, confidential information that rises to the level of a trade secret may receive special protections.  This typically requires two components:  (1) the information must derive independent economic value from not being generally known to, and (2) the holder must use reasonable efforts to maintain its secrecy.  This second requirement is of particular concern here.  A court could find that routine, or even one-time, submission of a company’s trade secrets into a public AI tool is evidence that such second requirement was not met, resulting in a loss of trade secret status.  This risk may be heightened if the company does not have an AI policy that prohibits such voluntary disclosures. Notably, using an enterprise or closed-box AI tool with proper contractual safeguards may help a company demonstrate “reasonable efforts” to maintain secrecy, but the analysis remains fact-specific, and the company should confirm the AI provider’s obligations in writing, ideally in a contract entered into before the first use of the enterprise system.

Your Contracts May Prohibit Use of AI

Before providing any third party data to an AI tool, public or private, a business should also consider its contractual duties with respect to such data.  For example, most confidentiality, joint-venture, and license agreements require a receiving party to protect the disclosing party's confidential information, and restrict disclosure to third parties.  The act of uploading customer data, a contract, or other confidential information received from a third party into an AI tool might violate those contractual obligations. 

Similarly, you may want to ensure that your own confidentiality agreements and NDAs prohibit parties to whom you disclose sensitive information from inputting any of your information into an AI tool, using it to train an AI large language model, or otherwise being put at risk of being inadvertently disclosed to the public and future users of the AI tool.  

Employees Often Create the Greatest Risk

Many AI-related disclosures occur because employees are trying to work more efficiently. They may upload an entire contract to ask AI for a summary or suggested changes, or paste an attorney's e-mail into AI to obtain a simpler explanation, all to help the employer.  However, without clear policies and training, well-intentioned employees may accidentally expose the company to unnecessary legal risk.  Even more, smart phones with built-in AI tools such as Apple Intelligence that automatically summarize emails or text messages for the user presents another version of this same risk.  Many businesses have yet to consider whether these indirect AI tools are considered open or closed box, and whether they increase the risk of unintended disclosure.

Practical Steps to Reduce Risk

Regardless of whether your company has officially adopted an enterprise AI system, know that your employees are using AI – if not with their work computers, then on their smartphones or personal devices.  Therefore, all companies should consider adopting an AI Usage Policy that includes, but is not limited to, the following:

• Prohibits employees from entering attorney communications or legal advice into public AI tools without express approval by senior management.
• Prohibits uploading confidential or proprietary information into any AI tool unless the platform has been specifically approved by senior management.
• Requires legal review before using AI with any trade secrets, protected health information (PHI) or other sensitive information.
• Evaluates whether the company’s AI tools are “closed-box” or enterprise-grade with contractual confidentiality protections, and restricts the use of public or consumer-grade AI for sensitive information.
• Reminds employees of the risk of AI hallucinations and tendencies, and not to rely on AI-generated content. Treat AI like a tool, not a replacement for common sense.
• Provides employee training on AI-related confidentiality risks.
• Includes periodic reminders to employees, at all levels, of these important AI policies. Reminders should be strategic, such as at times and places most likely to matter, and not just arbitrarily scheduled with a recurring calendar reminder.
• Requires due diligence on any AI platform before it is used with company information, including a review of the platform’s terms of service, privacy policy, and data-retention practices.
• Considers the difference between use of the more common AI providers (e.g. ChatGPT, Claude or Gemini), and new AI powered features that non-AI vendors are incorporating into their services and solutions (e.g. Apple Intelligence, CoPilot powered Google search results, etc.).

An AI Usage Policy should be reviewed and signed by all employees, contractors and agents as part of an onboarding process, and in regular training.

Key Takeaway

AI can be a valuable business tool, but it should not become an unintended method of disclosing privileged legal advice or confidential business information.  Before inputting confidential or legal documents, images, videos or data into any AI tool, businesses should understand the legal consequences.

The lesson from these early court decisions is clear that clients should exercise due diligence before using AI with any sensitive information. This means understanding whether the AI tool is a public, consumer-grade platform or a closed-box, enterprise version with contractual confidentiality protections. It also means being aware of the growing number of AI tools built specifically for lawyers and legal professionals, such as tools designed to maintain privilege and confidentiality by not retaining or training on user inputs. The landscape is evolving rapidly, and new lawyer-focused AI platforms continue to emerge.

Ultimately, the safest course is to speak with your attorney or compliance officer before inputting any confidential, trade secret, nonpublic, or potentially sensitive information into an AI system.