What the term means

What Is Confidential AI for a Law Firm?

Updated November 16, 2026

Short answer: "Confidential AI" means two different things. In computing, it is AI run inside hardware-isolated environments so that even the cloud operator cannot inspect the data while it is processed; in a law firm, it means an AI tool you can use on client information without breaching your duty of confidentiality. The first can help with the second, but a firm still has to check what the tool stores, who can see it and whether anyone trains on it.

What does "confidential AI" mean in computing?

It comes from confidential computing. The Confidential Computing Consortium, a Linux Foundation community, says "Confidential Computing protects data while it's in use", and describes hardware-enforced enclaves that "create isolated execution contexts where even the hypervisor and cloud provider can't inspect running workloads" (Confidential Computing Consortium).

Encryption has long covered data stored on disk and data moving across a network. Confidential computing aims at the gap in between: the moment the data is being worked on. "Confidential AI" applies that to the AI model's processing of your prompt.

That is a real protection, but a narrow one. It tells you who can look inside the machine while it runs. It does not tell you whether the vendor saves the conversation afterwards, keeps backups, or lets its staff read stored chats.

What makes an AI tool confidential enough for client information?

For a lawyer, confidentiality is a duty under Rule 1.6, not a technology. ABA Formal Opinion 512 says lawyers must "learn who has access to the information that the lawyer inputs into the tool" (p. 7) and check whether a tool "retains information submitted by the lawyer before and after the discontinuation of services" (p. 11) (ABA Formal Opinion 512). Five tests follow from that:

Test The question to ask What a vendor's own page can look like
Training Does anyone train AI models on what we enter? OpenAI: "We don't train our models on your organization's data by default." (OpenAI)
Storage What is kept after the answer comes back, and for how long? Anthropic, for Claude Team and Enterprise: "we retain your chats and coding sessions in the product" (Anthropic)
Access Who at the vendor, or its contractors, can read stored content? OpenAI, for ChatGPT Business: access by "authorized employees" and "specialized third-party contractors who are bound by confidentiality and security obligations, solely to review for abuse and misuse" (OpenAI)
Legal process Could a court order the vendor to preserve or produce it? Anything a vendor stores can be sought. See Can a law firm's AI chats be subpoenaed?
Location Where is the content processed and kept? Thomson Reuters, for CoCounsel: "We process all interactions with OpenAI GPT and Google Gemini in the U.S." (Thomson Reuters)

A tool that passes all five for your matters is confidential AI in the sense a law firm needs. A tool that is excellent on one and silent on the others is not there yet.

How is confidential AI different from private AI and zero data retention?

They overlap, and vendors use all three loosely:

  • Confidential AI is about protection while the AI is processing, and in legal use, about meeting Rule 1.6.
  • Private AI has no fixed meaning. Vendors often use it to say they do not train on your data and limit who can see it, which says little on its own about storage. Our explainer on private AI breaks the term down.
  • Zero data retention means nothing is kept after the answer is returned. The promise often covers only the model provider behind a product, not the product itself.

Which tools keep the least?

Three kinds, each with a cost:

  • Local AI on the firm's own computers. Nothing has to leave the office on the local-model path, but quality depends on the hardware and each lawyer usually works alone. Our local versus hosted page sets out the trade-offs.
  • Hosted tools that erase content. Proton says that after Lumo answers a query, "the data is erased" (Proton). At Faradex, each firm gets its own Workspace on dedicated infrastructure in the United States, and files and conversations are held in memory and are gone when the Workspace is reset.
  • Business plans with short retention. The large AI companies' business plans say they do not train on your data by default and offer admin controls, but they keep chat history by design. On some plans, such as ChatGPT Business, your admins can control how long it is kept.

This guide is general information, not legal advice. Check each vendor's current terms before you rely on them.

About Faradex. Faradex is a private AI workspace for small and mid-sized firms that handle confidential client work. Each firm gets its own Workspace on dedicated infrastructure in the United States. Files and conversations are held in memory and are gone when the Workspace is reset. AI model: Currently Opus via AWS Bedrock, more information is available on our website. AWS does not store prompts or responses, and model providers have no access to them. Pricing: from $120 a month, no seat minimum; see faradex.ai/pricing.