Cornerstone's headline number is the case count, but the exposure actually shows up in the loss ratio. Cornerstone attributes $385bn of the $529bn Disclosure Dollar Loss Index to AI filings [4][5], which works out to roughly $25.7bn of alleged loss per AI case against about $1.4bn for each of the other 102 filings, a gap of around eighteen to one [2]. That gap reflects which companies get sued rather than how AI cases are litigated: two filings alone carry $1.2tn of the $1.86tn Maximum Dollar Loss figure [6], an average of $600bn apiece, while the other 13 AI cases average under $8bn each [4]. The aggregate describes a handful of very large issuers, and a mid-cap reading it will misjudge both its own likely claim size and how ordinary its filing risk has become.
What the Cooley memorandum does not supply is the substance of the AI complaints. It sorts them by subject, seven on AI development and five on data centers, and stops there [2][16]. The shorthand that these are all overstatement-of-capability cases is therefore an inference from a category label rather than a finding in the record supplied, which is a real limit on how confidently any board can act on the trend this month.
Filings are not liability, and plaintiffs' firms often track stock drops rather than misstatements. The resolution data addresses that directly. NERA reported a slight decline in dismissals [7], and Cornerstone's H1 2026 median settlement of $20m sits about 54% above the 2017-2025 median of $13m, while the average, at $56.4m, is up about 21% on its own long-run mark [9][3]. The middle of the distribution moved further than the top, so the routine case is what became more expensive.
The tariff filings show the allegation pattern travelling. In those cases, Cornerstone reports, plaintiffs typically allege the company overstated its ability to manage the impact of tariffs, or understated how its own response would hurt the business [13]. The same overstatement theory applies whether the external condition is tariffs or AI capability. The exposure attaches to a competence claim about something moving faster than the disclosure cycle, and that is the trade-off leadership has to price: describing AI capability conservatively costs something real in a market that rewards the opposite, and describing it expansively produces an exhibit.
On timing, this is a quarter-scale problem resting on a short record. Annualizing the first half puts 2026 near double the 2025 AI total [5], and technology-sector filings went from nine in H2 2025 to 24, twice the semiannual average of 12 [11]. Two observations at a doubling do not establish a rate, and the consumer noncyclical sector still leads on volume with 44 filings [12]. What the record does establish is that the wording of an AI capability claim now has a measurable settlement range behind it, so disclosure committees are reviewing that wording this quarter, not waiting for next year's.
