By Wesley Page 

NOTE: Wesley Page is the Federal Public Defender for the Southern District of West Virginia. Prior to being head defender, Wes was a member of the CJA Panel and served as the CJA Panel District Representative. He now also serves as Chair of the Defender Automation Working Group (DAWG), a national working group that supports federal defenders and CJA panel attorneys in the effective and secure use of technology to ensure high‑quality legal representation. By way of background, DAWG advises the national Defender Services program by assisting in the program’s identification of technology needs, planning for future improvements, and sharing updates on tools and practices that enhance efficiency and casework for defenders and CJA panel attorneys alike.

Wes recently sent the following email to Federal Defenders and CJA panel representatives in response to the recent trend of “AI” protective orders. With his permission, we are reprinting his email here:

Recently, various Courts, judges, and the Government have attempted to govern the use of Artificial Intelligence (“AI”), including Generative AI (“GAI”) by Defenders in myriad ways.  Those ways, which we will collectively refer to as “AI Governance Efforts,” may take the form of local rules, standing orders, protective orders, or discovery letters and are seen to include, for example, requiring the disclosure of AI/GAI use to the Government and/or the Court, requiring defense counsel to obtain permission from the Government prior to using AI/GAI, and restricting or outright prohibiting the use of certain AI/GAI technologies in some or all instances.  It is our position that defense counsel should vigorously resist these efforts using all available legal and ethical means.

            Most importantly, the ABA Model Rules of Professional Conduct arguably require lawyers to understand AI/GAI and its uses and to consider how and whether those uses benefit their clients.  Beyond potentially impeding compliance with that ethical obligation, these AI Governance Efforts risk impinging on the attorney work product doctrine and attorney-client privilege (particularly with respect to counsel’s mental processes and litigation strategy vis-à-vis the choice of AI tools, search methodology, and analytical framework), undermine the efficient and effective use of new and emerging technologies to the detriment of our offices and clients, and create an uneven playing field where the Government’s use of AI/GAI goes unchecked.  With many attorneys already having implemented AI/GAI to some degree in their work, AI Governance Efforts seek to undermine much of the progress made in this area.

            We submit that a lawyer’s compliance with existing ethical obligations, including ABA Model Rules 2.1 (Independent Professional Judgment), 1.1 (Competence), 1.6, 1.9, and 1.18 (Confidentiality), 1.4 (Communication with the Client), 3.3 (Candor to the Tribunal), and 5.1 and 5.3 (Supervision), are sufficient to satisfy any legitimate concerns by the Government or judges regarding defense counsel’s use of AI/GAI.  Contrary to the primary concern of the AI Governance Efforts in protecting against unauthorized disclosure of sensitive information, the ethical and responsible use of AI/GAI in accord with those Model Rules will ensure that otherwise-confidential or sensitive information is not unreasonably compromised.  Moreover, many recent AI Governance Efforts are far too vague, broad, or overreaching to accomplish their intended purpose and instead stifle defense counsel’s legitimate use of AI/GAI in ways that are often unforeseen or unintended.  Specifically, many AI Governance Efforts extend to AI tools that pose no unreasonable risk of disclosure, including tools that perform rule-based or pattern-matching functions such as spell-check, keyword search, and document deduplication, as well as machine learning tools that classify or organize existing information without generating new content, such as technology-assisted review platforms used in eDiscovery. This overbreadth is itself a ground for resistance, and it is what the proposed language below is designed to correct.

            Fortunately, many attorneys to date have succeeded in opposing unreasonable AI Governance Efforts.  Should you encounter AI Governance Efforts, we are available to assist in your efforts to resist.  Recognizing that each situation is likely to present relatively unique circumstances and challenges, providing a one-size-fits-all solution is impracticable.

Most AI Governance Efforts, in attempting to address the concern about GAI, use the phrase “exposure to a third party” in ways that would apply to any cloud-based platform, including established eDiscovery review tools, because all such tools necessarily process data through vendor infrastructure to function. The critical distinction is not whether vendor infrastructure ever touched the material, but whether the vendor is contractually and technically barred from using or disclosing it for any purpose beyond the litigation at hand.  Similar to the use of a third-party vendor to make copies of voluminous materials or the use of cloud-based technology for email or document storage, there is no improper disclosure when AI is used, so long as the attorney has a reasonable basis to believe that that the material is secure against disclosure beyond that which is made to the vendor incidental to the service it is providing.   

With that distinction in mind, it may be helpful to provide suggested language that would address a legitimate concern of unauthorized downstream disclosure of sensitive information without needlessly restricting a lawyer’s ability to ethically and responsibly use AI/GAI: “Protected material should not be put into an AI tool unless the receiving party has a reasonable contractual or legal basis to conclude that the material will not be used to train models for unrelated users, will not be used for the vendor’s independent purposes, and will not be disclosed except as necessary to provide, secure, maintain, or support the tool for the receiving party’s permitted litigation use.”  To DAWG’s knowledge, several Defenders have used this or substantially similar language to help stave off oppressive AI Governance Efforts requiring permission or disclosure prior to AI/GAI use. The Standing Order regarding the Use of Artificial Intelligence published out of the Western District of North Carolina on July 29, 2026 may also be of assistance in educating stakeholders regarding an appropriate balance that recognizes the benefits of GAI while enforcing the duty of candor to the tribunal

            Insofar as the success of AI Government Efforts in any jurisdiction poses a threat to the ability to ethically and responsibly utilize AI/GAI elsewhere, it is critical that those efforts be met with the full force of resistance.  We hope that in recognition of that threat you will stand together with your colleagues in opposition to these AI Governance Efforts, utilizing our resources in whatever way you deem appropriate.