A federal court has now decided the question. A represented defendant's conversations with a consumer AI assistant were not privileged. Every chat is discoverable, subpoena-able, and admissible. Kovel wraps the same AI inside a real attorney relationship, so your legal thinking stays yours.
A defendant's conversations with Claude AI were seized by the FBI and ruled fully discoverable. A represented defendant's conversations with a consumer AI assistant were not privileged. Confidentiality failed independently of direction, because the provider's terms permitted access — so attorney direction alone would not have cured it on these facts.
Complete a brief intake. You're matched with a licensed attorney in your jurisdiction who formally engages you and directs your use of the platform.
Use the full power of frontier AI to research, analyze, and organize your legal thinking. Every session runs through the Anthropic API — not consumer products — and is not used for model training.
Your attorney relationship creates the legal shield. Documented attorney direction, confidential infrastructure, and a defensible privilege chain.
Bradley Heppner, indicted in the Southern District of New York on securities and wire fraud charges, used the consumer version of Claude to prepare roughly thirty-one documents outlining his defense strategy. He did it after receiving a grand jury subpoena and after it was clear he was the target of the investigation. When FBI agents executed a search warrant at his home in connection with his arrest, they seized documents and electronic devices, and those materials were among them.
Heppner asserted privilege over the documents and listed them on a privilege log. The Government moved for a ruling that they were protected by neither the attorney-client privilege nor the work product doctrine, and Judge Rakoff granted that motion. The court found the documents lacked at least two, and possibly all three, elements of the privilege. Claude is not an attorney, which the court held disposed of the claim on its own. The communications were not confidential, because the consumer privacy policy let the provider train on inputs and outputs and reserved a right to disclose them to third parties. And they were not made to obtain legal advice from Claude, which disclaims giving it. Confidentiality failed independently of direction, because the provider's terms permitted access — so attorney direction alone would not have cured it on these facts.
In the same opinion, in one hedged sentence, the court observed that had counsel directed the use, Claude “might arguably” have functioned as a lawyer's agent within the privilege, citing Kovel. That sentence is dictum, it addresses one of three defects, and it is not a safe harbour. It is also the only judicial language anyone has on what a defensible arrangement would look like, and it is what Kovel is built around.
Securities fraud, wire fraud, conspiracy, false statements to auditors, and falsifying corporate records. Unsealed on November 4.
Documents and electronic devices seized in connection with the arrest. Among them, roughly thirty-one documents memorializing his conversations with Claude.
Not a defense motion to suppress. Heppner had asserted privilege and logged the documents; the Government asked the court to rule they were not protected.
After argument at the pretrial conference. The written memorandum setting out the reasons follows on February 17.
"Had counsel directed Heppner to use Claude, Claude might arguably be said to have functioned in a manner akin to a highly trained professional who may act as a lawyer's agent within the protection of the attorney-client privilege."
Every draft, strategy note, and private reflection typed into a public AI is potentially an exhibit — unless it's wrapped in the Kovel framework.
Vendor contracts, NDAs, partnership agreements, SaaS terms. Stress-test clauses before you sign — with privilege intact.
Wrongful termination, wage theft, workplace disputes, regulatory exposure. Build your timeline and strategy without creating discoverable records.
Pre-charge investigation prep, grand jury targets, compliance exposure, whistleblower strategy. Highest-stakes research under the strongest privilege.
Give your AI the context it needs to be useful. Upload case law, contracts, strategy playbooks, internal memos, and prior filings to create a custom knowledge base that informs every AI interaction.
Firms set context at the firm level (standard playbooks, jurisdiction-specific precedent, boilerplate language) or at the individual client level (case-specific documents, opposing counsel filings, deposition transcripts). The AI references this material in its research, making output immediately relevant to your actual matters.
Documents and conversations stay within the privilege-protected infrastructure. Nothing is used for model training.
Standard operating procedures, jurisdiction-specific case law, preferred motion templates, and firm-wide legal research libraries.
Upload contracts, filings, correspondence, depositions, and strategy documents. AI responses reference your actual case materials.
Retrieval-augmented generation within the privilege boundary. Documents indexed locally, never sent to third parties, never used for training.
All model calls route through the Anthropic API, not consumer-facing products. Per Anthropic's Terms of Service, API inputs and outputs are not used to train models.
Prompts and responses exist for the life of your session. You decide what persists to your Law Library; nothing else is stored.
All traffic uses TLS encryption in flight. Law Library documents and conversation data are encrypted at rest on our infrastructure.
All infrastructure hosted in US-East with no cross-border replication. No subprocessors outside the US.
Every session records the directing attorney, the engagement scope, and the matter — exportable as an evidentiary log in any privilege challenge.
If Kovel receives legal process affecting your matter, we will make reasonable efforts to notify the directing attorney before responding, consistent with applicable law.
Join the Kovel attorney network and receive matched clients who need lightweight advisory coverage. You set the scope. We handle the platform.
Clients come pre-qualified. You issue the engagement letter, direct their use of the AI portal, and are available for periodic consultation. Build your Law Library with firm-specific knowledge to make the AI more valuable for your clients.
Pre-qualified clients matched to your jurisdiction and practice area. No marketing spend required.
Advisory retainer model. Direct AI use and periodic consultation. No full-service obligations.
Offer branded AI research with your firm's playbooks, precedent, and templates baked into every interaction.
Updated engagement letter templates, privilege documentation, and compliance workflows built in.
No existing counsel? We match you with a vetted attorney in our network. They sign the directing engagement, and your research begins — typically within 24 hours.
Already working with counsel? Invite them directly. They sign a one-page Kovel agreement, become your supervising attorney, and gain a read-only view of your research sessions.
Privilege in this context rests on the Kovel doctrine — United States v. Kovel, 296 F.2d 918 (2d Cir. 1961) — which extends attorney-client privilege to third-party agents engaged by counsel to facilitate legal advice. The canonical example is an accountant retained by a tax attorney.
Every Kovel account includes a real engagement letter with a licensed attorney who directs your use of the AI platform as part of legal advisory services. Direction is necessary but not sufficient: Direction and a confidential channel are both required, and consumer tools cannot supply the second. Our infrastructure supplies the second half, and the contemporaneous record is what makes the claim defensible in a privilege challenge.
Your prompts and the AI's responses exist for the duration of your active session and are discarded at session end unless you explicitly save them to your Law Library. We run every model call through the Anthropic API tier — per Anthropic's Terms of Service, API inputs and outputs are not used to train models.
Anything you do save is encrypted at rest and is only accessible to you and your matched attorney.
A real attorney is slow and expensive. Kovel gives you the research and thinking assistant of an AI — with privilege protection in place — so you can show up to your (human) attorney session having already organized the facts, drafts, and questions. It reduces the billable hours you owe while increasing the quality of the conversation.
Yes. Your Individual plan includes quarterly demand-letter service and on-demand consultation windows with your matched attorney. Some matters also qualify for full-scope representation with additional fees — your attorney will tell you when that's appropriate.
Your attorney gives you legal advice. Kovel is the research environment they direct you to use. The AI outputs are research and analysis, not advice — and the platform explicitly routes high-stakes decisions back to the directing attorney.
We're rolling out jurisdiction by jurisdiction, starting with NY, CA, TX, FL, IL, and DC. If you sign up and we don't yet have an attorney in your state, we'll hold your spot, pair you with a courtesy attorney licensed in a neighboring jurisdiction where permissible, and upgrade you when local counsel is available.
Monthly plans cancel anytime from your account page. Your engagement letter includes a termination clause and a defined process for retrieving or deleting your Law Library materials.
Yes — and in fact the Kovel doctrine itself originated in a criminal tax case. The key is how you use Kovel relative to the stage of the matter.
Pre-charge (investigation, grand jury, target letter, worry about exposure): Kovel is ideal. You get a privileged research environment under general counsel's direction while you figure out whether and when to engage defense counsel.
Post-charge (charged, arraigned, in active defense): engage a criminal defense attorney as a specialist on that matter in Kovel. That attorney directs your AI research within the matter, and Kovel serves as their research infrastructure. The defense attorney runs the defense; Kovel supports it.
What Kovel doesn't do: we don't replace defense counsel, and our general-counsel master engagement doesn't cover courtroom representation. If your matter is approaching charges or already charged, you'll see a prompt inside the portal to request attorney review and match to a criminal defense specialist.
Every prompt you type into a public AI is a potential exhibit. Kovel makes sure it stays yours.
Request beta access