AI Searches Ruled Discoverable in Federal Criminal Proceeding
A federal court in New York published its rationale for a recent ruling in a criminal case that the defendant’s searches on a generative artificial intelligence (AI) platform are not protected from discovery by federal prosecutors. The ruling is the first in the nation to address whether AI search histories connected to a case are protected by either the attorney-client privilege or the work product doctrine. The court ruled they are not, and companies should note the ruling’s implications for their use of the ever-growing and accessible technology.
Key Points
- Unguided AI use is discoverable
- Training minimizes risk
Background
The case involves allegations against a corporate executive, Bradley Heppner, of fraud, conspiracy, falsifying records, and lying to auditors. A federal grand jury returned an indictment against Heppner, alleging he defrauded investors out of more than $150 million. In connection with Heppner’s arrest, federal agents seized documents and electronic devices. Heppner’s attorney later said those items contained communications between Heppner and Claude, an AI chatbot like ChatGPT, Grok, and other commercial large language models (LLMs), about the government’s investigation into his alleged criminal activity.
Heppner conducted the searches and prepared reports outlining defense strategies without direction from an attorney. The government wanted to see those searches, and Heppner sought to keep them sealed as attorney-client privileged materials, work product excluded from discovery, or both. Heppner’s counsel argued the materials were not discoverable because Heppner created them after it was clear he was the target of the investigation and in anticipation of his indictment.
The Court’s Analysis
The court disagreed. To be privileged, the court said communications must be between a client and his or her attorney, be both intended as and in fact kept confidential, and be made for the purpose of obtaining or providing legal advice. According to the court, Heppner’s communications on Claude failed at least two of the requirements and perhaps all three.
First, Claude is not an attorney. Thus, Heppner’s communications could not meet the first prong of the test. On the position espoused by commentators that whether an AI platform “is an attorney is irrelevant because the user’s inputs are not communications” but more “akin to the use of other Internet-based software” like cloud-based word processing applications, the court found all privileges require “a trusting human relationship” with a licensed individual who owes a fiduciary duty to the sender and is subject to discipline. Neither Claude nor any of its analogous competitors qualify.
Second, the court found the communications were not confidential because Claude is a third-party whose privacy policy Heppner consented to states that the platform’s operator collects data on user input and platform output to train the LLM’s algorithm. In addition, Claude’s owner reserves the right to disclose user searches with third parties, including the government. At bottom, the court found that Claude users do not have a substantial privacy interest in their conversations with the LLM.
Last, the court found that the absence of direction by an attorney to Heppner on his conduct of the searches removed the communications from the definition of those made to obtain legal advice. When queried on whether it could give legal advice, Claude responded, “I’m not a lawyer and can’t provide formal legal advice or recommendations.” In addition, the court dropped a footnote stating that, to the extent Heppner shared privileged information with Claude, doing so waived the privilege as to that information.
The court made short work of Heppner’s argument that his search histories constituted work product shielded from discovery. It did so by finding that, even assuming the inputs and reports were prepared in anticipation of litigation, they were not prepared by or at the direction of counsel. Moreover, according to the court, Heppner’s input and reports failed to reflect his counsel’s strategy at the time they were created and therefore fell outside the definition of work product.
Understand and Assess Your Risks
The proliferation of AI presents enormous opportunities in the transportation sector, where leveraging data can mean the difference between success and failure. Leaders, including in-house counsel, must foster cultures that not only capitalize on those opportunities but also respect the pitfalls they present. That includes training on the use of AI in a manner that preserves both confidential and privileged information. Although the Heppner case involves criminal charges, no meaningful distinction in the privilege and work-product analyses will likely save civil defendants from similar rulings under like circumstances. Policies drafted in coordination with counsel are a must to maximize strategic advantage while preserving the sanctity of sensitive information.
If you have any questions or want to discuss your approach to managing your company’s AI risks, feel free to reach out to Scopelitis attorneys Shannon Cohen or Jim Spolyar.
News from Scopelitis is intended as a report to our clients and friends on developments affecting the transportation industry. The published material does not constitute an exhaustive legal study and should not be regarded or relied upon as individual legal advice or opinion.
AI Searches Ruled Discoverable in Federal Criminal Proceeding
A federal court in New York published its rationale for a recent ruling in a criminal case that the defendant’s searches on a generative artificial intelligence (AI) platform are not protected from discovery by federal prosecutors. The ruling is the first in the nation to address whether AI search histories connected to a case are protected by either the attorney-client privilege or the work product doctrine. The court ruled they are not, and companies should note the ruling’s implications for their use of the ever-growing and accessible technology.
Key Points
- Unguided AI use is discoverable
- Training minimizes risk
Background
The case involves allegations against a corporate executive, Bradley Heppner, of fraud, conspiracy, falsifying records, and lying to auditors. A federal grand jury returned an indictment against Heppner, alleging he defrauded investors out of more than $150 million. In connection with Heppner’s arrest, federal agents seized documents and electronic devices. Heppner’s attorney later said those items contained communications between Heppner and Claude, an AI chatbot like ChatGPT, Grok, and other commercial large language models (LLMs), about the government’s investigation into his alleged criminal activity.
Heppner conducted the searches and prepared reports outlining defense strategies without direction from an attorney. The government wanted to see those searches, and Heppner sought to keep them sealed as attorney-client privileged materials, work product excluded from discovery, or both. Heppner’s counsel argued the materials were not discoverable because Heppner created them after it was clear he was the target of the investigation and in anticipation of his indictment.
The Court’s Analysis
The court disagreed. To be privileged, the court said communications must be between a client and his or her attorney, be both intended as and in fact kept confidential, and be made for the purpose of obtaining or providing legal advice. According to the court, Heppner’s communications on Claude failed at least two of the requirements and perhaps all three.
First, Claude is not an attorney. Thus, Heppner’s communications could not meet the first prong of the test. On the position espoused by commentators that whether an AI platform “is an attorney is irrelevant because the user’s inputs are not communications” but more “akin to the use of other Internet-based software” like cloud-based word processing applications, the court found all privileges require “a trusting human relationship” with a licensed individual who owes a fiduciary duty to the sender and is subject to discipline. Neither Claude nor any of its analogous competitors qualify.
Second, the court found the communications were not confidential because Claude is a third-party whose privacy policy Heppner consented to states that the platform’s operator collects data on user input and platform output to train the LLM’s algorithm. In addition, Claude’s owner reserves the right to disclose user searches with third parties, including the government. At bottom, the court found that Claude users do not have a substantial privacy interest in their conversations with the LLM.
Last, the court found that the absence of direction by an attorney to Heppner on his conduct of the searches removed the communications from the definition of those made to obtain legal advice. When queried on whether it could give legal advice, Claude responded, “I’m not a lawyer and can’t provide formal legal advice or recommendations.” In addition, the court dropped a footnote stating that, to the extent Heppner shared privileged information with Claude, doing so waived the privilege as to that information.
The court made short work of Heppner’s argument that his search histories constituted work product shielded from discovery. It did so by finding that, even assuming the inputs and reports were prepared in anticipation of litigation, they were not prepared by or at the direction of counsel. Moreover, according to the court, Heppner’s input and reports failed to reflect his counsel’s strategy at the time they were created and therefore fell outside the definition of work product.
Understand and Assess Your Risks
The proliferation of AI presents enormous opportunities in the transportation sector, where leveraging data can mean the difference between success and failure. Leaders, including in-house counsel, must foster cultures that not only capitalize on those opportunities but also respect the pitfalls they present. That includes training on the use of AI in a manner that preserves both confidential and privileged information. Although the Heppner case involves criminal charges, no meaningful distinction in the privilege and work-product analyses will likely save civil defendants from similar rulings under like circumstances. Policies drafted in coordination with counsel are a must to maximize strategic advantage while preserving the sanctity of sensitive information.
If you have any questions or want to discuss your approach to managing your company’s AI risks, feel free to reach out to Scopelitis attorneys Shannon Cohen or Jim Spolyar.
News from Scopelitis is intended as a report to our clients and friends on developments affecting the transportation industry. The published material does not constitute an exhaustive legal study and should not be regarded or relied upon as individual legal advice or opinion.