J.S. Held's Q3 AI disputes report: the legal fight over AI moves from how models are built to what they do

Global consulting firm J.S. Held released the Q3 2026 update to its AI Disputes Monitor on Wednesday, and the numbers sketch a legal front that has moved downstream: 42 new matters added in the quarter, 468 AI-related cases tracked in total, and a growing share focused not on the data that trained AI systems but on what those systems did once they were put to work.

From how it's built to how it behaves

The Monitor's Office of the Chief Intellectual Property Officer reports that the 42 additions brought the tracked dataset to 468 cases, including 115 actions filed in 2026 through September 30. Early AI litigation concentrated heavily on the materials used to train models and whether developers had authorization to obtain, copy, or use protected content — and those questions remain active. But the Q3 matters increasingly ask how systems behaved after deployment, including their outputs, product safeguards, automated decisions, corporate disclosures, and board oversight.

"The litigation is expanding from the inputs used to build AI systems to the consequences of putting those systems into operation," said James E. Malackowski, J.S. Held's Chief Intellectual Property Officer and co-founder of Ocean Tomo, part of J.S. Held. The evidence, he said, may now need to explain "not only where the data came from, but how the system behaved, what controls were in place," and what the organization claimed about its capabilities.

The evidence itself is changing

The expansion changes what a court will ask to see. Legal and intellectual-property disputes remain the largest individual market segment in the update, but litigation exposure is spreading across the whole AI value chain — model developers, technology platforms, companies deploying AI in their operations, public institutions, corporate officers and directors, and users alleged to have circumvented safety controls.

A dispute may now require examination of model versions, prompts, outputs, system logs, integration records, testing evidence, data provenance, human-review procedures, governance documentation, market impact, and economic loss. As the analysis puts it, "across product safety, privacy, automated decisions, securities, and governance, a recurring question is whether the organization can reconstruct how the system operated and demonstrate that appropriate controls were functioning." Policies alone, the Q3 analysis warns, will not answer that question.

Two ways to get disclosure wrong

The Q3 additions also point to two distinct forms of AI-related disclosure risk, per the Monitor's findings. The first concerns whether an organization's promoted AI capabilities, integrations, partnerships, or commercial benefits were substantiated — the classic "AI washing" allegation, where claims about AI sophistication or value allegedly exceed the supporting evidence.

The second runs the other way: whether a company adequately disclosed how AI was disrupting its own business — its products, pricing models, customer behavior, or growth prospects. Here the allegation is not necessarily that the company's AI capabilities were fictitious, but that it underrepresented AI's competitive effect on its established business. Disclosure risk, in other words, runs both ways: overstating what AI can do, and understating what AI may do to the business — a distinction that matters for directors, officers, investors, insurers, and counsel alike.

Why it matters

J.S. Held launched the Monitor earlier this year to track what it describes as a fast-growing litigation landscape: AI-related litigation growing at a 56.5% five-year compound annual rate, with 93% of cases filed in the United States and concentrations in California, New York, and Illinois, according to the firm's launch data. The Q3 update is the latest evidence that the dispute wave follows the deployment wave.

For organizations building or buying AI, the practical takeaway is concrete: training-data hygiene is no longer the whole risk surface. Logging, testing records, escalation procedures, decision documentation, and proof of meaningful human oversight are becoming the evidence that matters when something goes wrong in production. AI systems, as the analysis notes, now function simultaneously as technical products, economic assets, and sources of evidence — and courts are starting to treat them as all three.