Court decides information shared by non-experts with AI tools not protected by …

In a recent court ruling, U.S. District Judge Jed S. Rakoff determined that information shared with an artificial intelligence (AI) tool by an individual without the involvement of legal counsel does not fall under the protection of attorney-client or work-product privileges. The judgement was made in a case involving a defendant, Heppner, who utilized an AI tool called Claude, developed by Anthropic, to conduct searches related to an ongoing government investigation in securities and wire fraud. Notably, Heppner input data acquired from his defense counsel at Quinn Emanuel into the AI tool and subsequently shared the generated documents with his legal team.

When the FBI confiscated Heppner’s devices, his defense counsel argued that the 31 documents shared were privileged. However, the government contended otherwise, asserting that the communications with the AI tool were not protected by attorney-client privilege or the work-product doctrine. Judge Rakoff supported the government’s motion, highlighting that no foundation for an attorney-client privilege existed in this scenario. The court’s ruling was based on specific arguments put forth by the government regarding this case.

Firstly, the court acknowledged that there is no attorney-client privilege when an individual seeks guidance from a commercial AI tool, like Claude, as these tools are not licensed attorneys and hence cannot establish an attorney-client relationship. Additionally, the terms of service and privacy policy of these AI tools stipulate that the provided data may be used for training purposes and can be disclosed to authorities.

Secondly, the court emphasized that the transmission of search queries and results to counsel after using an AI tool does not retroactively create attorney-client privilege. According to established legal principles, previously non-privileged documents do not attain privileged status solely because they are later shared with an attorney. The court found no distinction between sharing AI-generated responses with counsel and conducting a Google search for information.

Moreover, the court determined that the information gathered through the AI tool did not qualify for protection under the qualified work-product doctrine. Since the AI interactions were not directed by counsel in anticipation of litigation, but were independently conducted by Heppner prior to sharing them with his legal team, they did not meet the criteria for work-product protection.

This ruling underscores the potential risks associated with the use of AI tools and serves as a reminder to exercise caution while engaging with them. The decision establishes a precedent that independent interactions with AI tools, especially at the consumer level, may lead to the disclosure of sensitive or privileged information. It is advisable for individuals and organizations to be mindful of what data is shared with AI tools, given the possibility of losing privilege protection.

In addition to the ruling’s direct implications, there are broader considerations to bear in mind. For instance, if a client seeks legal advice from counsel and subsequently shares that advice with an AI tool, there is a potential risk of waiving the attorney-client privilege. If similar circumstances arise in the future, courts are likely to adopt similar stances on the disclosure of privileged communications to AI tools, emphasizing the importance of safeguarding confidential information.