“Your Client Is Talking to ChatGPT About Their Case. After ‘Heppner,’ That’s a Discovery Problem.” —

• “You have a client right now who is typing their case details into ChatGPT. A personal injury client describing their injuries and uploading medical records. An employment client recounting what HR said in the termination meeting. Both recapping what you told them on the phone last week. They think it is private. It is not. And after United States v. Heppner, opposing counsel has a federal court opinion telling them how to use that against your client.”
• “Most of the commentary on this ruling has come from Big Law defense shops advising corporate clients on internal AI policies. That tracks. But the real exposure lands on the plaintiff’s side. Your clients are the ones awake at 2 a.m., anxious, asking a chatbot what their case is worth. Defense attorneys are not losing sleep over Heppner. They are sharpening discovery requests.”
• “On Feb. 10, Judge Jed Rakoff ruled in the Southern District of New York that 31 documents a criminal defendant created using a consumer version of Claude were not privileged. The defendant had fed information from his attorneys into the chatbot, generated defense strategy reports, and later shared them with counsel. The court held the materials failed on at least two, and likely all three, elements of privilege: an AI tool is not an attorney, the consumer platform’s privacy policy negated any expectation of confidentiality, and the defendant acted without counsel’s direction.”
• “Rakoff did leave one door open. If counsel had directed the client to use the AI tool, the court suggested it might function like a third-party expert working under the attorney’s supervision. That question remains unanswered.”
• “Here is where it gets dangerous for plaintiff firms specifically. Heppner did not just create new documents with AI. He fed privileged attorney communications into the platform. Information his lawyers gave him went into Claude as prompts. Judge Rakoff’s opinion suggests this may have waived privilege over the original attorney-client communications, not just the AI outputs.”
• “Think about what that looks like in your practice. Your client takes your case evaluation, your strategy notes, your assessment of liability, and types it into ChatGPT to ‘understand it better.’ Under Heppner’s reasoning, the defense can argue the privilege over your advice just evaporated. That is not a hypothetical edge case. That is a Tuesday.”
• “The discovery angle cuts both ways, and plaintiff attorneys should be thinking offensively here. Corporate defendants and their employees are using AI tools to summarize internal investigations, evaluate liability exposure, and draft talking points. If those conversations happened on consumer platforms without counsel’s direction, the same Heppner logic could make them discoverable. Plaintiff firms fight against well-resourced corporate legal teams every day. This is a new tool in the belt.”
• “What to Do About It. Update your engagement letters. Add explicit language that anything a client types into a consumer AI platform about their case may be discoverable. This belongs right next to your social media warning.”
• “Have the conversation at intake. Ask directly: are you using ChatGPT, Claude, or any AI tool to research your case or your injuries? Most clients will not volunteer this. You need to ask and you need to explain why it matters.”
• “Document attorney direction. If you want a client using AI for case preparation, put your instructions in writing. Specify the tool, specify the purpose. Heppner drew a bright line between client-initiated and attorney-directed use. Stay on the right side of it.”
• “Expect AI-targeted discovery. Heppner was a criminal privilege ruling, but the weight of legal commentary says the reasoning extends to civil discovery. Defense counsel reads the same case law you do. Requests for production targeting AI usage are coming.”
• “Heppner applied old privilege principles to new technology. The conversation you need to be having with every new client is simple: if you are using AI to think about your case, tell me now, and stop putting case details into consumer chatbots. That one conversation could save the case.”

“Silent Ransom Group leaked another big law firm: Orrick, Herrington & Sutcliffe” —

• “Jones Day wasn’t the only big law firm to recently fall prey to threat actors variously known as Silent Ransom Group, Luna Moth, Chatty Spider, or UNC3753. DataBreaches will refer to them as the Silent Ransom Group (‘SRG’).*”
• “In January, SRG gained access to the law firm of Orrick, Herrington & Sutcliffe LLP (‘Orrick’). In terms of gross revenue, Orrick is not as large as Jones Day. Still, it has over 25 offices across the United States, Europe, and Asia, and its gross revenue exceeded $1.5 billion in 2025.”
• “Orrick’s name may sound familiar because in March 2023, it was the victim of a data breach that affected 461,000 people. A consolidated class-action lawsuit was settled in 2024, with Orrick not making any admission of guilt but agreeing to create an $8M non-reversionary fund as part of the settlement.”
• “Having dealt with the costly 2023 breach, would they be eager to avoid another class action by paying SRG in the hope that the breach never becomes widely known? Or would they decide not to pay at all and weather any storm?”
• “In communications with SRG, DataBreaches asked several questions, one of which was, ‘Did the fact that Orrick had settled that class-action for $8M have any effect on you targeting them? Did you think that a big firm that has just made a big settlement would be more likely to pay to avoid a second scandal or big lawsuit?’ They replied:”
  • “As We can see, this organization is not learning from its mistakes. Typically, when dealing with firms from the top 100 law firms, they all, without exception, recognize the seriousness of the situation and make the only correct decision.”
  • “They offered $1,000,000 to resolve this matter, but that was far less than the amount requested, and We decided to publish their details. This was the first top-100 law firm to offer such a meager sum.”
• “SRG informed DataBreaches that it first gained access to Orrick on or about January 20, 2026, and remained in its network for about a week. The attack did not involve any malware, as SRG never deploys malware.”
• “On February 6, 2026, a representative from Orrick appeared in chat and negotiations began. SRG shared the chat logs with DataBreaches, but only after negotiations had failed.”
• “From statements made in the chat by both parties, it seems that SRG contacted employees and Orrick’s clients at times. The negotiations appeared calm on both sides, with Orrick’s representative saying at one point:”
  • “I can tell you are experienced in this subject matter, which means you must also know thet [sic] lawsuits happen and information becomes public no matter what the outcome is here. It does not mean we’re unwilling to find an agreement, but we want to speak honestly about what a deal does and does not accomplish. I hope that is OK with you. We do not agree that [redacted] million is a reasonable expectation for purchasing the word of a stranger. It does not ease any of the other costs we will face from this. If you have a different number in mind that is more realistic, we are happy to review it and begin a productive dialogue.”
• “SRG responded by reviewing some facts and listing the names of some of the clients whose private information was now in SRG’s hands. Orrick did not come back with what SRG would consider commensurate with the scope and volume of the data involved. On February 18, Orrick offered $225,000.00 and subsequently flat-out told SRG that while they were open to making a deal with SRG, ‘there is no 7 figure deal to be had here.’”
• “On February 23, SRG added Orrick to its leak site and leaked all of the data. Orrick’s negotiator noted the development and offered SRG a final $400,000.00 offer if the post was removed immediately. It was removed for a while, but SRG rejected Orrick’s subsequent $1M final offer and re-leaked it all.”
• “Some might think that any hack-and-leak gang would be happy to get $1M and would take the money and run. But what SRG did is similar to what DataBreaches has seen the Daixin Group and others do in other incidents: they refuse an offer that seems way too low so that future victims will learn the group will leak rather than accept an offer they do not consider reasonable.”
• “DataBreaches’ inspection of the data tranche identified many files that appeared to be confidential. DataBreaches will not focus on the more confidential information, but will provide a sense of how little file security there was for some files. In one of several index files in the leak, 77 filenames included ‘CONFIDENTIAL’ in their names. All of those files were in plaintext with no password protection. Some of them contained confidential files relating to litigation. Some contained confidential and sensitive employee information.”
• “As part of its settlement in 2024 IN RE: ORRICK, HERRINGTON & SUTCLIFFE, LLP DATA BREACH LITIGATION, Orrick also agreed to improve its security:”
  • “As further consideration to the Settlement Class, Orrick has agreed to make numerous business practices changes relating to data security. Orrick has also confirmed that, as a direct result of Plaintiffs’ filing of the Action, Orrick has already implemented several improvements to its data security. Ex. 1, §8.1. These enhancements include improving its detection and response tools, enhancing its continuous vulnerability scanning at both the network and application levels, deploying additional endpoint detection and response software, and with the help of an industry leading cybersecurity vendor, performing additional 24/7 network managed detection and response.”
• “As the FBI noted in its private industry notification last year, law firms are frequent and lucrative targets due to the sensitivity of the information they store, and SRG, in particular, has been targeting them since 2023. Jones Day and Orrick are not the only law firms SRG has hit — and leaked — recently. DataBreaches may have more on this in the near future.”