Frontier-AI Litigation Exposure for GCs

This page is not legal advice, and nothing below establishes a legal standard. It is a plain-language survey of three real, currently checkable court and regulatory matters, offered so in-house counsel have a shared vocabulary for discussing frontier-AI liability exposure — not a prediction of how any of them will end, and not a substitute for jurisdiction-specific advice from a lawyer admitted to practice where your organization operates. The law in this area is genuinely unsettled: as of this writing, no US court has established a frontier-AI liability standard, and two of the three matters below remain open and contested. Where a case is Canadian, that is stated explicitly, because it carries no precedential weight in any US jurisdiction.

Last verified: September 20, 2026. The short answer for a general counsel skimming this page: one small-claims tribunal in British Columbia has actually held a company liable for its chatbot’s output; one US federal age-discrimination claim against an AI hiring-screening vendor has cleared a procedural hurdle without being decided on the merits; and one set of wrongful-death and product-liability claims against a companion-AI company is still at the pleading stage in multiple states, alongside a single state licensing action that is not a lawsuit at all. None of that is a pattern of established liability — it is three separate fact patterns, at three different stages, in three different legal postures, and treating them as one trend would overstate what any of them actually holds. NIKOLAI, CASRAI’s own independent, unendorsed frontier-AI-safety dictionary, gives this page one useful thing: a discipline for classifying incidents by their underlying mechanism rather than by which legal theory a plaintiff’s lawyer reached for first — more on where that helps, and where it doesn’t yet reach, below.

Key Facts at a Glance

Moffatt v. Air Canada, 2024 BCCRT 149 — decided February 14, 2024, by the British Columbia Civil Resolution Tribunal, a Canadian small-claims-level administrative tribunal. Air Canada was held liable for its chatbot’s negligent misrepresentation. This is not a US case and sets no US precedent.
Mobley v. Workday — an active federal case in the Northern District of California. In May 2025, the court allowed Age Discrimination in Employment Act (ADEA) collective-action claims to proceed against Workday as an AI hiring-screening vendor and ordered disclosure of a client list. No liability has been found; the case is ongoing.
Garcia v. Character Technologies — filed October 2024 in Florida federal court, alleging wrongful death and product-liability claims against Character.AI. Related suits are pending in Colorado and Texas. All remain unresolved.
Pennsylvania Department of State action, filed May 2026 — a licensing and regulatory enforcement matter, not a lawsuit and not brought by a state Attorney General, over Character.AI chatbot personas presenting as licensed medical professionals.
No pattern of state AG enforcement is established by anything on this page. Exactly one regulatory action is confirmed here, and it is a licensing-board matter, not an Attorney General suit.

Case 1: The Chatbot That Made a Promise Its Employer Wouldn’t Keep

In late 2022, a customer named Jake Moffatt used Air Canada’s website chatbot to ask about bereavement fares after his grandmother’s death. The chatbot told him he could book a flight at the regular fare and apply for a bereavement discount retroactively, within 90 days. That was wrong — Air Canada’s actual policy required the bereavement request before travel — and when Moffatt filed for the retroactive discount, Air Canada refused, pointing to a separate page on its own site with the correct policy.

Moffatt brought a small claim before the British Columbia Civil Resolution Tribunal, a provincial administrative body that resolves disputes under a set claims threshold; it is not a court of general jurisdiction, and its decisions are not binding outside British Columbia, let alone outside Canada. Tribunal member Christopher C. Rivers found Air Canada liable for negligent misrepresentation and awarded Moffatt CA$812.02 plus fees and pre-judgment interest — not thousands of dollars, and not punitive damages; the remedy was the difference between what Moffatt paid and what the correct bereavement fare would have cost.

The detail that made this decision widely discussed was Air Canada’s own defense: it argued the chatbot was “a separate legal entity that is responsible for its own actions.” The tribunal rejected that argument directly, reasoning that the chatbot was part of Air Canada’s website and that Air Canada is responsible for the information on its own site regardless of whether a human or a tool produced it. That reasoning — a company cannot disclaim its own customer-facing tool as a separate actor — is the part of this decision most often cited outside Canada. But it was decided by a small-claims tribunal applying British Columbia law to a BC$812 dispute, not by any court, and it has no formal precedential force in the United States or in any other Canadian province. Treat it as a data point about how one adjudicator reasoned about one narrow defense, not as a holding any US general counsel can cite as binding authority.

Case 2: An AI Vendor’s Screening Tool, Tested but Not Yet Decided

Mobley v. Workday is a different kind of case entirely: not a single customer’s small claim, but a federal collective action testing whether a company that builds AI-powered hiring-screening software can be treated as an “agent” of the employers who use it, and therefore be directly liable under federal anti-discrimination law. Lead plaintiff Derek Mobley alleges Workday’s applicant-screening tools disproportionately filtered out older applicants, among other protected-class allegations, across the many employers who use Workday’s platform to screen candidates.

The case is proceeding in the US District Court for the Northern District of California. In May 2025, the court allowed Mobley’s Age Discrimination in Employment Act (ADEA) claims to proceed as a collective action — meaning other applicants who believe they were screened out by Workday’s tools under similar circumstances can join the case — and the court’s order included discovery obligations for Workday, reported to include disclosure of client information relevant to identifying the pool of affected applicants. This procedural ruling is not a finding of liability. Allowing a collective action to proceed answers a narrow question — are these claims similar enough to litigate together — not the underlying question of whether Workday’s tools actually discriminated against anyone. Workday disputes the claims and continues to litigate the case. As of this writing it remains open, with no judgment on the merits.

What makes this case worth a general counsel’s attention regardless of outcome is the theory being tested, not a result: whether an AI vendor supplying a screening tool to many employers can itself be a defendant under employment-discrimination law, separately from the employers who deploy the tool. No US court has yet resolved that theory one way or the other. A ruling allowing the theory to be tested in discovery is meaningfully different from a ruling establishing that the theory is correct.

Case 3: Unresolved Claims, Three States, and One Regulator

Garcia v. Character Technologies is the most serious and the least resolved of the three matters on this page. Filed in October 2024 in a Florida federal court, the suit brings wrongful-death and product-liability claims against Character Technologies, the company behind the Character.AI companion-chatbot platform, arising from the death of a minor after extended use of the platform’s chatbot characters. The complaint frames the claims in product-liability terms — alleging the platform was designed without adequate safeguards for vulnerable users — rather than solely in terms of the chatbot’s individual statements, a different legal theory than the negligent-misrepresentation claim in the Air Canada matter above.

This is not an isolated filing. Related suits raising similar wrongful-death and harm-to-minors allegations against the same company are pending in Colorado and Texas, arising from separate incidents. None of these cases has been decided on the merits. They remain at early procedural stages — motions, discovery, and, in at least one matter, disputes over whether the company’s terms of service can compel arbitration — not trial or settlement. Nothing in any of these filings has been adjudicated as fact, and the allegations in a complaint are, procedurally, only allegations until a court rules otherwise.

Separately — and this distinction matters enough to state plainly — the Pennsylvania Department of State and State Board of Medicine filed a regulatory enforcement action in May 2026 alleging that certain Character.AI chatbot personas held themselves out as licensed psychiatrists and other medical professionals, in potential violation of Pennsylvania’s Medical Practice Act. This is a licensing and regulatory matter brought by a state professional-licensing body, not a lawsuit and not an action brought by a state Attorney General. It is the only state regulatory action of any kind confirmed for this page. Readers should not infer from it that a wave of state AG enforcement against frontier-AI companies is underway — that pattern is not established by anything verified here, and asserting it would overstate what one licensing-board filing actually shows.

Where NIKOLAI Fits

NIKOLAI is CASRAI’s own independent, unendorsed reference dictionary of frontier-AI-safety vocabulary — not a legal standard, not a compliance framework, and not affiliated with, endorsed by, or consulted by any court, regulator, or company named on this page. It has no bearing on how any of the three matters above will be decided, and nothing in it should be read as legal guidance. What it offers a general counsel tracking this space is narrower and more modest: a shared vocabulary for talking about AI-related incidents in terms of what actually happened, rather than only in terms of which legal cause of action a plaintiff filed.

NIKOLAI’s N7 — Incidents track includes an element called Incident Type (N7.3), currently tagged proposed in NIKOLAI’s schema. Verified directly against the live element page on September 20, 2026, its definition reads, verbatim: “Incident type is a controlled value classifying an Incident by mechanism and severity class, so that similar events can be compared across developers.” NIKOLAI’s own editorial note is unusually direct about what this element is: “This is CASRAI’s own proposed definition, not a definition any named organisation has agreed to” — built by synthesizing candidate categories including model exfiltration, unauthorized access, loss-of-control events, deceptive-subversion incidents, materialized catastrophic harms, and lower-severity precursor or near-miss events, because, in NIKOLAI’s own words, “no two sources currently share one enumeration.”

Here is the honest limit worth naming rather than glossing over: none of the three fact patterns on this page maps cleanly onto N7.3’s current candidate categories. A chatbot’s negligent misrepresentation about a fare policy, an AI screening tool’s alleged age-discriminatory filtering, and a companion chatbot’s alleged role in a minor’s death are not “model exfiltration” or “loss-of-control” events in the sense N7.3 was built to classify — NIKOLAI’s incident-type vocabulary was synthesized primarily from frontier-lab safety-incident reporting (the kind covered by frameworks like California SB 53), not from consumer-facing product-liability or employment-discrimination fact patterns. That gap is itself informative for a general counsel’s purposes: the vocabulary the AI-safety field has been building classifies a different slice of harm than the one actually showing up in court dockets right now. What N7.3 does usefully model, independent of its specific categories, is the underlying discipline — classify an incident by its causal mechanism and severity first, and only then map it onto whichever legal theories that mechanism happens to support — which is a reasonable habit for a legal team tracking a fast-moving docket across jurisdictions, even where NIKOLAI’s own enumeration doesn’t yet reach the fact pattern in front of you.

Frequently Asked Questions

Has any US court ruled that an AI company is liable for its chatbot’s statements?

Not as of this writing. Moffatt v. Air Canada is the clearest liability finding involving a chatbot, but it was decided by a Canadian small-claims tribunal under British Columbia law and has no precedential force in any US court. The two active US matters described on this page — Mobley v. Workday and Garcia v. Character Technologies — remain unresolved, with no liability finding on the merits in either.

Does Mobley v. Workday mean AI hiring vendors are now liable for discrimination?

No. The May 2025 ruling allowed ADEA claims to proceed as a collective action, which is a procedural decision about whether similar claims can be litigated together — it is not a finding that Workday’s tools discriminated against anyone. The underlying liability question remains open, and Workday continues to contest the claims.

Is there a state Attorney General crackdown on frontier-AI companies?

Nothing verified for this page supports that characterization. The one confirmed state regulatory action here — the Pennsylvania Department of State and State Board of Medicine’s May 2026 filing against Character.AI — is a professional-licensing enforcement matter, not a lawsuit, and it was not brought by a state Attorney General. Treating one licensing action as evidence of a broader AG enforcement pattern would overstate what has actually happened.

Is the Air Canada chatbot decision binding precedent in the United States?

No. It is a February 14, 2024 decision of the British Columbia Civil Resolution Tribunal, an administrative small-claims body, applying Canadian and provincial law to an approximately CA$812 dispute. It is frequently cited as illustrative of how one adjudicator reasoned about a “separate legal entity” defense, but it has no formal precedential weight in any US jurisdiction.

Is this page legal advice?

No. It is a general, educational survey of publicly reported case developments, current as of the verification date above, prepared for general awareness. It does not account for your organization’s specific facts, jurisdiction, or risk posture, and it is not a substitute for advice from a lawyer admitted to practice where you operate. The status of every matter described here can and likely will change after this page’s last-verified date.

General Counsel’s Guide to Frontier-AI Litigation Exposure