Pennsylvania Is Warning the Bar: AI Liability Has Arrived
The first reported state action against Character.AI fits an old pattern: once harm is foreseeable, the question becomes duty, not novelty.
An AI-assisted editorial, reviewed by a human before publishing. It reasons over our own tracker data (and cited context) — a point of view, not legal advice.
You can feel the shift in the way a case gets discussed before anyone has even filed a brief. First it is a curiosity, then it is a warning, and then a state starts acting as if the harm was always part of the design. Pennsylvania’s first reported action against Character.AI belongs in that third bucket. For a working lawyer, that matters more than the branding on the product.
Once the harm is foreseeable, regulators stop asking whether it is just a tool.
The old lesson from the archive
We learned this lesson years ago in the sanction cases that made AI fiction expensive. In Mata v. Avianca, Inc., Schwartz filed a brief with six ChatGPT-fabricated cases and the court fined him and co-counsel $5,000 for misleading it. In Whiting v. City of Athens, Tennessee, the Sixth Circuit sanctioned two Tennessee lawyers for appellate briefs with over two dozen fake or misrepresented citations and facts. In Shahid v. Esaam, the Georgia Court of Appeals struck a divorce order that cited bogus cases and sanctioned attorney Diana Lynch.
What changed
Those were courtroom consequences for bad filing hygiene. Pennsylvania points to something broader: the state is no longer waiting for a lawyer to hand in fake citations before it treats AI harm as actionable. The working rule is becoming plain. Once a technology produces foreseeable harm, regulators stop asking whether it is just a tool and start asking who had a duty to prevent the damage. That is the same logic behind the National Center for State Courts’ AI work, which says courts are issuing guidance and resources as they respond to AI and AI-related issues, and it is the same logic behind Florida’s 2026 bill on companion AI chatbots, which would require age verification, notices, and let the Department of Legal Affairs bring actions and collect civil penalties.
Why the bar should care now
This is not a niche consumer-tech story. HaystackID says generative AI has moved from experimentation to standard use in eDiscovery, with hundreds of customers using it for review and privilege and more than 100 million documents analyzed. Microsoft’s Corporate, External, and Legal Affairs organization, about 2,000 lawyers and compliance professionals, is also going to use Harvey AI for legal and compliance work. On the firm side, Latham & Watkins has launched an AI Academy to train lawyers on AI and points to deal work tied to Apollo, Broadcom, and CoreWeave. AI is already inside the practice, the department, and the court file. Pennsylvania just tells you the next place liability will land.
The point lawyers keep missing
The temptation is to treat each new event as a separate category: sanctions here, in-house adoption there, regulation somewhere else. That is the wrong reading. The pattern is simpler and harsher. Once the harm is foreseeable, institutions move from advice to enforcement. Once enforcement starts, the lawyer who thought the machine was only a tool discovers that duty has a habit of attaching to the person who used it, signed it, or failed to control it.