For attorney investigators, Artificial Intelligence (AI) should not be shunned but rather embraced with a deliberate understanding that foundational privilege principles remain paramount. As Gorev Ahuja of Oppenheimer Investigations Group articulates, a judicious approach—one that prioritizes confidentiality safeguards, employs approved platforms, and exercises sound professional judgment—will position legal professionals to harness AI’s efficiencies while robustly preserving the crucial protections required for workplace investigations. The increasing prevalence of alarming headlines such as "Lawyers Sanctioned Over Fake AI Citations," "AI Hallucinations Reach the Courtroom," and "Courts Warn: Attorneys Cannot Delegate Judgment to Chatbots" underscores a growing judicial and ethical scrutiny of AI’s integration into legal practice.
Courts and regulatory bodies have unequivocally stated that while AI offers valuable tools, it does not absolve legal professionals of their fundamental duties of competence, accuracy, and independent judgment. As AI becomes more deeply embedded in the investigative process, attorneys face the critical challenge of understanding how its application impacts the vital tenets of confidentiality, attorney-client privilege, and the work-product doctrine.
The Distinctive Role of Investigations in Employment Law
Investigations are a cornerstone of employment law, with employers frequently relying on legal counsel to scrutinize allegations of discrimination, sexual harassment, retaliation, fraud, and various other forms of workplace misconduct. When these investigations are conducted under the direction of counsel and with the explicit purpose of rendering legal advice, they are typically shielded by the attorney-client privilege and the work-product doctrine. These legal doctrines are designed to foster an environment conducive to thorough and candid fact-gathering, free from the apprehension of subsequent disclosure that could compromise legal strategies or client confidences.
The transformative power of AI is now significantly reshaping the investigative landscape. AI tools are demonstrating remarkable capabilities in summarizing lengthy interviews, meticulously organizing vast datasets of documents, identifying subtle inconsistencies, constructing detailed timelines, and even generating preliminary draft reports. While these efficiencies are undeniably substantial, they are not without considerable legal risks that demand careful consideration.
Central to these emerging concerns are questions such as: What are the implications when privileged witness statements are input into an AI platform? Are the prompts used to query these systems, and their subsequent outputs, subject to discovery? Furthermore, does the utilization of AI inherently undermine the confidentiality necessary to maintain the integrity of privilege? These complex questions emerge at the confluence of two rapidly evolving domains: investigative practices and generative AI. Although many of these intricate issues have yet to be definitively adjudicated by the courts, existing legal principles governing privilege offer a clear and crucial starting point for analysis.
Understanding Privilege and Work Product in Investigations
Both state and federal laws provide robust protections for attorney-client communications and attorney work product. The attorney-client privilege, a cornerstone of legal practice, safeguards confidential communications exchanged for the purpose of obtaining or delivering legal advice. Distinct from this, the work-product doctrine offers protection to materials prepared by attorneys in anticipation of litigation or in connection with their legal representation.
Within the context of investigations, these protections are of paramount importance. Effective legal counsel and strategic advice are often contingent upon a comprehensive and unhindered development of factual evidence. Courts have consistently recognized that, under appropriate circumstances, an investigation directed by an attorney is not a separate or ancillary activity but an integral component of providing legal advice itself.
A landmark decision by the U.S. Supreme Court, Upjohn Co. v. United States, established a crucial precedent: communications between corporate employees and in-house counsel, made for the express purpose of assisting counsel in providing legal advice to the corporation, are protected under the attorney-client privilege. This principle has been further elaborated and reinforced by state courts. In California, for instance, the Court of Appeal’s decision in City of Petaluma v. Superior Court held that an outside attorney’s investigative report concerning alleged misconduct was protected by both the attorney-client privilege and the work-product doctrine. This ruling was significant because it affirmed protection even when the attorney’s role was primarily fact-gathering, provided the investigation was undertaken to facilitate the provision of legal services.
Additional judicial pronouncements have solidified the strength of these protections within California’s legal framework. For example, Costco Wholesale Corp. v. Superior Court emphasized the broad scope of the attorney-client privilege once it is deemed applicable. Concurrently, Coito v. Superior Court recognized substantial work-product protections for attorney interview materials that encapsulate an attorney’s impressions, conclusions, opinions, or legal theories.
The overarching takeaway from these foundational cases is that investigations conducted under attorney direction are entitled to significant legal protections. However, as with any analysis of privilege, the availability and extent of these protections are invariably dependent on the specific factual matrix of each case, with a critical element being the demonstrable maintenance of confidentiality.
AI and the Delicate Nature of Confidentiality
The very existence of legal privilege hinges on the preservation of confidentiality. This requirement becomes considerably more complex when AI is introduced into the investigative workflow. Historically, attorneys relied on human intermediaries, such as paralegals, external consultants, and eDiscovery vendors, to assist in data analysis and processing. Courts have generally held that engaging these agents does not constitute a waiver of privilege, provided their involvement is reasonably necessary to facilitate the delivery of legal advice.
AI, however, introduces a distinct set of considerations. A significant number of AI platforms operate on cloud-based systems managed by third-party providers. Depending on the specific platform’s architecture and its governing terms of use, information uploaded into the system may be stored, processed, or even retained by the provider to enhance its AI models. This raises a critical legal question: Does the act of entering privileged information into an AI system constitute an unauthorized disclosure to a third party, thereby potentially waiving privilege?
While definitive judicial answers remain nascent, emerging case law offers valuable insights into how traditional privilege principles might be applied in this rapidly evolving technological context.
In the federal court case of Tremblay v. OpenAI, Inc., a judge grappled with the question of whether AI prompts themselves are discoverable. The court acknowledged that prompts crafted by attorneys can reveal intricate legal theories, strategic planning, and the attorney’s mental impressions. Consequently, the court suggested that such attorney-generated prompts could potentially qualify as opinion work product, a category of work product that enjoys near-absolute protection from discovery. A subsequent ruling echoed this sentiment. In Concord Music Group, Inc. v. Anthropic PBC, a court similarly recognized work-product protections for attorney-crafted investigative prompts and their associated outputs under the specific circumstances presented in that case.
For legal investigators, these rulings are of considerable significance. Prompts designed to elicit summaries of witness testimony, assess credibility, organize evidentiary exhibits, or formulate investigative findings often embody the investigator’s strategic thinking, mental impressions, and approach to a case. Under the reasoning in Tremblay, such prompts may indeed be eligible for work-product protection. However, it is crucial to understand that this protection is not automatic. Courts will likely conduct a thorough examination of the content, context, and stated purpose of these prompts to ascertain whether they genuinely reflect an attorney’s or investigator’s thought processes.
A more cautionary precedent emerged from United States v. Heppner. In this case, a defendant utilized a publicly accessible AI platform to generate documents discussing legal theories and potential defenses. When these AI-generated materials were subsequently seized, the court dismissed claims of privilege and work-product protection. The court’s reasoning was straightforward: the communications were not with legal counsel, they were voluntarily disclosed to a third-party provider, and they were not created at the direction of counsel.
The implications of the Heppner decision are substantial. If applied broadly, the use of a consumer-grade AI platform to process privileged investigation materials could present a compelling argument for the waiver of both confidentiality and, consequently, privilege.
It is important to note that the Heppner case involved a public platform and did not directly address the scenario where an attorney is performing legal work on behalf of a client. Therefore, it left open the critical question of whether an attorney employing enterprise-level AI systems, which often come with contractual confidentiality agreements, should be treated differently.
Offering a contrasting perspective, the case of Warner v. Gilbarco, Inc. presented a different factual scenario and yielded a different outcome. In this employment discrimination lawsuit, a self-represented plaintiff employed a generative AI tool to assist in preparing litigation-related documents. The defendants subsequently filed a motion to compel production of all materials associated with the plaintiff’s AI usage, arguing that any applicable protections had been waived. The court denied the motion. In its analysis of work product, the court drew a critical distinction: while attorney-client privilege can be waived by disclosure to a third party, work-product protection is only waived by disclosure to an adversary or through conduct that is likely to place the material in an adversary’s hands. The court reasoned that AI tools are "tools, not persons," and therefore, entering litigation materials into such a tool did not constitute disclosure to an adversary. Furthermore, the court determined that the defendants’ request improperly sought the plaintiff’s internal analyses and mental impressions, which were deemed protected work product because she had prepared them in anticipation of litigation.
The most recent development in this evolving area, Morgan v. V2X Inc., involved a pro se plaintiff in another employment discrimination matter. In this instance, the defendant sought to modify an existing protective order to incorporate AI-specific restrictions and to compel the disclosure of the specific AI platform used by the plaintiff. The plaintiff argued that her choice of AI tool was protected work product. The court granted the defendant’s motion in part.
The Morgan court explicitly declined to adopt the approach taken in Heppner, concluding that intermediary access to data does not automatically result in a waiver of protections. However, the court did rule that the identity of the AI tool itself was not protected and ordered its disclosure. Crucially, the court also established an AI-specific protective order, mandating that no party may input confidential information into an AI platform unless the provider’s contract explicitly prohibits the use of such inputs for training purposes or their disclosure to third parties.
Practical Guidance for Attorney Investigators
Courts are increasingly applying established principles of privilege and work-product doctrine to AI-assisted legal work. In doing so, they continue to pose many of the same fundamental questions that have long guided privilege analyses: Was there a qualifying relationship in place? Was confidentiality rigorously maintained? Were the materials prepared in anticipation of litigation? And do they reflect the mental processes of a party, attorney, or investigator?
AI introduces a novel factual context for these enduring legal questions, rather than an entirely new set of rules. Nevertheless, numerous critical questions remain unsettled, and the outcomes of future cases are likely to be heavily influenced by the specific factual circumstances, including the precise nature of the AI platform employed and the robustness of the confidentiality protections implemented.
AI undoubtedly possesses the capacity to significantly enhance efficiency in areas such as document review, evidence organization, chronology development, and preliminary drafting. However, it is imperative to remember that investigators retain ultimate responsibility for critical tasks such as evaluating witness credibility, weighing the significance of evidence, resolving conflicting information within the investigative record, and formulating final findings. AI-generated outputs should always undergo independent review and verification before being integrated into any investigative process or official report. While AI can serve as a valuable assistive tool, it cannot, and should not, substitute for the exercise of professional judgment.
In the realm of workplace investigations, the developing case law strongly suggests that attorneys and organizations must proactively implement thoughtful strategies when utilizing AI to support their investigative efforts. The primary objective must be to preserve applicable privilege and work-product protections. Several key principles emerge from this evolving legal landscape, offering practical guidance for navigating this complex terrain:
-
Platform Selection is Critical: Prioritize AI platforms that offer robust contractual assurances regarding data confidentiality, security, and the prohibition of data usage for training or third-party sharing. Enterprise-grade solutions with clearly defined service agreements are generally preferable to public or consumer-facing tools.
-
Maintain Confidentiality Rigorously: Treat all information input into AI systems with the same level of confidentiality as any other privileged or sensitive document. Avoid using AI for sensitive matters on platforms that do not offer verifiable data protection.
-
Document AI Usage Deliberately: Keep meticulous records of how AI tools are used in investigations. This includes detailing the prompts used, the outputs generated, and how those outputs were reviewed and utilized. This documentation will be crucial in demonstrating that AI was used as a tool to assist, not replace, legal judgment.
-
Understand Prompt Engineering as Legal Strategy: Recognize that the prompts used to interact with AI can reflect legal strategy and mental impressions. Treat these prompts with the same care as other work-product materials.
-
Independent Verification is Non-Negotiable: Never rely solely on AI-generated summaries, analyses, or drafts. Always conduct independent review, fact-checking, and critical evaluation of AI outputs to ensure accuracy and completeness.
-
Consult with Legal Counsel: When in doubt about the implications of using a particular AI tool or about maintaining privilege, seek advice from experienced legal counsel specializing in e-discovery, technology law, and privilege issues.
By embracing AI with a strategic, informed, and privilege-conscious approach, attorney investigators can leverage its powerful capabilities to enhance their effectiveness while upholding the essential legal protections that underpin their critical work in employment law and beyond. The journey through this evolving landscape requires vigilance, adaptability, and an unwavering commitment to established legal principles.
