Cloud AI services create privilege waiver risk through third-party disclosure. On-premises AI hardware keeps client data inside the firm, under firm control, with zero external transmission.
Cloud-based AI services create privilege waiver risk because client data transmitted to third-party servers may constitute voluntary disclosure to a non-privileged party. On-premises AI hardware eliminates this risk entirely. Data never leaves the firm's physical control, preserving the confidentiality element required under ABA Model Rule 1.6 and consistent with ABA Formal Opinion 477R guidance on technology competence.
Attorney-client privilege protects confidential communications between a lawyer and client made for the purpose of obtaining legal advice. The privilege is waived when the communication is voluntarily disclosed to a third party. When a lawyer pastes client facts into a cloud AI prompt, that data is transmitted to servers owned by OpenAI, Anthropic, Google, or another provider. The provider's employees, systems, and subprocessors may access the data for model training, safety review, abuse monitoring, or legal compliance.
Even providers that contractually promise not to use inputs for training still process the data on their infrastructure. The transmission itself is a disclosure to a non-privileged party. Whether a court would find waiver depends on jurisdiction, the sensitivity of the information, and the specific provider's data handling practices - but the risk exists every time client data leaves the firm's control.
The Supreme Court's holding in Upjohn Co. v. United States (449 U.S. 383, 1981) established that the privilege belongs to the client and can be waived by any voluntary disclosure - including careless handling by the attorney. FRE 502 provides some protection against inadvertent disclosure, but relying on FRE 502 as a safety net for routine cloud AI usage is a strategy built on hope rather than policy.
ABA Model Rule 1.6(a) prohibits lawyers from revealing information relating to the representation of a client without informed consent. Rule 1.6(c) requires lawyers to make reasonable efforts to prevent unauthorized disclosure. ABA Formal Opinion 477R (2017) extends this to electronic communications, requiring lawyers to assess the sensitivity of information before choosing a transmission method and to use reasonable safeguards proportional to the sensitivity.
Comment 8 to Rule 1.1 (amended 2012) adds a duty of technology competence: lawyers must stay abreast of changes in technology relevant to their practice. Using a cloud AI service to process sensitive client data without understanding the provider's data retention, training, and access policies is difficult to reconcile with this duty.
Discovery risk compounds the problem. Under FRCP Rule 34, opposing counsel can request production of electronically stored information. If a firm used cloud AI to process case-related documents, the prompts, responses, and any data retained by the provider may be discoverable. This creates a second vector of exposure beyond the privilege waiver issue.
On-premises AI hardware processes all queries on local NVIDIA GPUs inside the law firm's physical facility. No client data is transmitted to any external server. There is no third-party disclosure, no API endpoint for opposing counsel to subpoena, no provider data retention policy to evaluate, and no terms-of-service training clause to parse.
Island Mountain's Summit Base system ($75-85K) ships with open-source models running locally: DeepSeek V4-Flash for fast inference, Llama 3.3 70B for deep reasoning, and a buyer-selected DeepSeek R1 70B Distill or Qwen 2.5 72B. Open WebUI provides a familiar chat interface with individual user accounts and per-session query logs that remain under the firm's control.
The firm's existing confidentiality infrastructure - physical access controls, network security, employee NDAs - extends naturally to cover the AI system. No new third-party relationship is created. No new BAA, DPA, or cloud provider contract is needed.
Potentially. The transmission constitutes disclosure to a third party. Whether a court finds waiver depends on the jurisdiction, the sensitivity of the data, and the provider's data handling practices. The safest position is to never transmit privileged information to a cloud AI service.
Enterprise agreements may reduce but do not eliminate the risk. The data still leaves the firm's control and is processed on third-party infrastructure. A contractual promise not to train on inputs does not change the fact that a disclosure occurred. And contracts can be breached, amended, or overridden by legal process.
Yes. Under FRCP Rule 34, electronically stored information held by the firm or its service providers is potentially discoverable. With on-premises AI, query logs remain on hardware the firm controls. With cloud AI, logs may exist on servers the firm cannot fully audit or delete.
Summary: Cloud-based AI services create attorney-client privilege waiver risk by transmitting client data to third-party servers. ABA Model Rule 1.6 requires confidentiality, and ABA Formal Opinion 477R requires reasonable safeguards for electronic communications. On-premises AI hardware eliminates the third-party disclosure vector entirely. Island Mountain's Summit Series servers process all queries locally on NVIDIA H100/H200 GPUs, starting at $75,000.
Learn more: Law Firms AI Infrastructure | Attorney-Client Privilege & Cloud AI | OpenAI Discovery Risk for Law Firms
Talk to Island Mountain about on-premises AI that keeps client data inside your building. No cloud. No waiver risk. No third-party exposure.
Request a QuoteOr call directly: 1-801-609-1130