Abstract
Legal technology companies have deployed large language model-based products that provide specific, jurisdiction-sensitive legal guidance to consumers without a licensed attorney in the loop. The products occupy a studied ambiguity: they are labeled “legal information” rather than “legal advice,” a distinction that has historically shielded legal publishers from unauthorized practice of law (UPL) liability. That framing is now under pressure from state bar associations, consumer advocates, and some courts, and the gap between what these systems do in practice and what existing regulation covers is substantial.
The Information-Advice Line and Why It Is Breaking Down
The traditional boundary between legal information (general explanations of law) and legal advice (application of law to a specific person’s facts) was workable when the information side was static text in a pamphlet. LLM-based systems ask about the user’s facts, reason over jurisdiction and procedural posture, generate customized documents, and recommend specific actions. Products from DoNotPay’s successors and a range of legaltech startups now routinely cross what any working attorney would recognize as the advice line, while their terms of service disclaim any attorney-client relationship. California’s State Bar issued a notice in late 2024 flagging this pattern; the ABA’s working group on AI and legal services published preliminary guidance that acknowledged the definition problem without resolving it.
Liability Exposure for Harmed Users
The ethical stakes are not abstract. A consumer who follows AI-generated guidance on a landlord-tenant dispute, an immigration application, or a debt collection response may take actions that harm their legal position, and they may have no recourse. Attorney malpractice doctrine does not attach because no attorney was involved. Product liability for software advice is unsettled. The FTC’s authority over deceptive practices is a plausible but untested hook. Several state legislatures introduced bills in 2025 requiring disclosure when legal documents are AI-generated; none of those bills included enforcement mechanisms or private rights of action.
What Responsible Deployment Looks Like
The responsible path, in the view of bar association ethics opinions that have begun to trickle out, involves three elements: mandatory disclosure that the system is not an attorney, clear jurisdictional scope limiting (not just a disclaimer but an actual hard cutoff when the question requires jurisdiction-specific advice the system cannot safely give), and a warm handoff to a licensed attorney for anything touching contested facts or adversarial proceedings. Systems that implement none of these while describing themselves as “your personal legal assistant” are operating in a gap the regulatory apparatus has not yet closed, and that gap creates real harm in the populations most likely to rely on free or low-cost legal tools.