Attorney-Client Privilege in the Age of Agency

A Legal Doctrine Meets Its Technological Moment

Attorney-client privilege is one of the oldest and most important protections in the legal system. Communications between lawyers and their clients are protected from disclosure, enabling the candid conversations that effective legal representation requires. Without this protection, clients would hesitate to share the full truth with their attorneys, and justice would suffer.

For centuries, this doctrine has been refined around a simple model: a human client talks to a human lawyer. But what happens when AI agents enter the conversation? When a client's AI assistant communicates with a law firm's AI system? When neither party in an exchange is actually human?

These questions are no longer theoretical. They're being asked in courts right now. And the answers will reshape how legal services are delivered and how startups engage with counsel.

The Traditional Framework

To understand what's changing, we need to understand what's been stable. Attorney-client privilege traditionally requires four elements:

First, there must be a communication. This is broadly defined—written, oral, electronic, it doesn't matter. What matters is that information is being transmitted.

Second, the communication must be between privileged parties. This typically means the client and their attorney, though it extends to certain agents of each—paralegals, legal assistants, and others reasonably necessary for the representation.

Third, the communication must be made in confidence. Privilege doesn't protect statements made in public or shared with third parties who aren't part of the privileged relationship.

Fourth, the communication must be for the purpose of seeking or providing legal advice. Business advice, even from a lawyer, isn't privileged. The legal nature of the communication is essential.

This framework has accommodated many technological changes. Email, cloud storage, video conferencing—courts have generally found that privilege extends to new communication technologies, so long as the underlying relationship and intent remain privileged.

Where AI Breaks the Model

AI agents create novel challenges for each element of this framework.

Is there a "communication" when an AI system generates output? The human may never see the intermediate steps. If your AI assistant queries a legal database, processes the results, and synthesizes advice—was there a privileged communication, or just computation?

Who are the "privileged parties" when AI is involved? Is an AI assistant an agent of the client, like a paralegal is an agent of the attorney? Courts have historically required that agents be "reasonably necessary" for the representation. Is an AI reasonably necessary, or a convenience? Does it matter whether the AI is operated by the client, the attorney, or a third-party service provider?

Is there "confidentiality" when AI systems process communications? If your message to your attorney is processed by an AI system operated by a cloud provider, has confidentiality been breached? What if that AI system was trained on similar communications? What if it could theoretically retrieve or reproduce portions of the communication later?

And what about the "legal advice" requirement? AI systems often blend legal information with other kinds of analysis. If an AI agent provides recommendations that are partly legal and partly business, which portions are privileged?

The Emerging Questions

Several specific scenarios are working their way through courts and legal ethics opinions:

Client-side AI. When a client uses an AI assistant to communicate with their lawyer, is that communication privileged? Most analysis suggests yes, so long as the AI is under the client's control and the communication remains confidential. But questions remain about AI systems that transmit data to third-party providers, retain communications for training purposes, or could potentially be compelled to disclose their contents.

Attorney-side AI. When a law firm uses AI to process client communications, does this break privilege? Under the traditional framework, privilege extends to the attorney's agents who are reasonably necessary for the representation. Most ethics opinions have concluded that AI tools fall within this category—but typically with significant conditions around confidentiality, data handling, and vendor agreements.

AI-to-AI communication. The hardest case is when both sides deploy AI agents that communicate with each other. A client's AI assistant might negotiate with a law firm's intake AI before any human sees the conversation. Is this privileged? The traditional framework has no clear answer because it presumes human parties on both sides.

The Corporate Context

For corporate clients, these questions are even more complex. Corporate privilege already involves nuances around which communications are protected—the "control group" test, the "subject matter" test, the Upjohn factors for corporate investigations. Adding AI to this mix creates additional uncertainties.

When a company deploys AI agents that interact with legal counsel, who is the "client" for privilege purposes? The corporation as an entity? The employees who interact with the AI? The executives who directed its deployment?

If AI systems are used in internal investigations—a common scenario for corporate legal departments—how do the privilege protections for investigation work products extend to AI-generated analysis? Can opposing parties compel disclosure of AI outputs that synthesize privileged and non-privileged information?

Practical Guidance for Founders

While courts and bar associations work through these questions, founders engaging legal counsel need to navigate the current uncertainty. Several principles can help:

Control matters. Privilege is strongest when the AI system is under the direct control of a privileged party—the client or the attorney. Using your own AI assistant to communicate with counsel is lower risk than routing communications through a third-party AI service.

Confidentiality provisions matter more than ever. If you're using AI services that process legal communications, scrutinize the confidentiality provisions in those services' terms. Do they retain your data? Use it for training? Could they be compelled to disclose it? Weaker confidentiality provisions create stronger arguments that privilege was waived.

Document the legal purpose. Communications that are clearly seeking or providing legal advice are more likely to be protected than ambiguous ones. When AI is involved, the legal purpose should be explicit and documented.

Ask your lawyer. Seriously. These are questions with significant consequences and evolving answers. Your legal counsel should be advising you on how AI affects your privileged communications, including how their firm uses AI in handling your matters.

The Bigger Picture

This isn't just a legal technicality. Privilege exists because it enables justice. Clients need to be fully candid with their lawyers. If AI deployment creates uncertainty about whether that candor will be protected, clients may hesitate—and legal outcomes will suffer.

For startups building AI products, there's also an opportunity. Tools and services that demonstrably preserve privilege—through technical architecture, contractual commitments, and compliance certifications—will be more valuable to clients engaging with legal counsel. Privilege-preserving AI may become a meaningful differentiator.

The legal system adapts slowly, but it does adapt. The core principles of attorney-client privilege will likely extend to AI-mediated communications, with appropriate safeguards. But during this transition period, founders need to be more thoughtful—not less—about how AI affects their most sensitive communications.