Res Ipsa Machina This Week — July 3, 2026

The first appellate argument on fair use in AI training, a state AI act repealed before it ever took effect, and two courts escalating hallucination sanctions well past the fine-and-scold stage. For an inaugural issue, the docket cooperated.

1. Third Circuit hears the first appellate argument on fair use in AI training

   LawSites · Copyright

   On June 11 a panel of Judges Restrepo, Montgomery-Reeves, and Bove heard Thomson Reuters v. ROSS Intelligence, the appeal from the District of Delaware's ruling that training a legal-research tool on Westlaw headnotes was not fair use. Argument concentrated on factors one and four — transformativeness and market harm — with ROSS stressing that its non-generative system returned judicial text rather than substitute headnotes, and Thomson Reuters framing the case as pure market substitution. Whatever the panel writes will be the first appellate word on AI training and fair use, and every generative AI case in the pipeline will spend next year distinguishing it.

2. Colorado repeals and replaces its AI Act before it ever takes effect

   Norton Rose Fulbright · Regulation

   SB 24-205, the first comprehensive state AI statute, was due to take effect June 30. It never did: Governor Polis signed SB 26-189 on May 14, swapping the algorithmic-discrimination duty of care and impact assessments for a disclosure-centered regime governing automated decision-making technology, effective January 1, 2027. The retreat is hard to separate from the litigation backdrop — xAI's constitutional challenge, joined by DOJ in the first federal intervention against a state AI law, had already produced a suspension of enforcement. The lesson other statehouses will draw is that first-mover AI mandates now carry federal litigation risk priced in.

3. Federal judge removes all four lawyers from a case over AI hallucinations

   Mississippi Free Press · Practice of Law

   Judge Sharion Aycock of the Northern District of Mississippi found that counsel on both sides of a contract dispute had filed AI-fabricated authorities, and on June 8 removed all four attorneys from the case — barring two from practicing in the district for two years. The order is notable for rejecting the learning-curve defense outright: three years after Mata, ignorance of hallucination risk no longer mitigates. The sanctions ceiling is moving from fines toward disqualification, and the pro hac vice angle should worry every out-of-state litigator.

4. New York appellate court sanctions a lawyer — and his firm — over fake citations

   Minnesota Lawyer · Practice of Law

   The Appellate Division, Second Department ordered attorney Michael Sanders to pay 8,000 dollars and his firm 2,500 dollars to the state Lawyers' Fund for Client Protection after an appellate brief cited nonexistent decisions and fabricated quotations that unraveled at oral argument. Two details matter beyond the fine: the firm was sanctioned as an entity for failing to enforce its own verification policy, and the panel expressly teed up possible referral to the grievance committee. Hallucination discipline is becoming an institutional-liability problem, not just an individual one.

5. Albany closes its session with an AI bill blitz — now on Hochul's desk

   Transparency Coalition · Regulation

   New York legislators ended the 2026 session on June 1 by passing a cluster of AI measures: a chatbot-safety bill for minors with a private right of action (S 9051), the AI Training Data Transparency Act (A 6578), the FAIR News Act on AI-generated news content (S 8451), plus surveillance-pricing and data-center measures. Governor Hochul has until December 31 to sign, and the federal preemption campaign will loom over every signature decision. If the training-data bill becomes law, New York joins California in forcing disclosure fights that the copyright litigation has so far failed to resolve.

6. Illinois bets that third-party audits are the durable form of AI safety law

   Akerman · Regulation

   SB 315, passed with near-unanimous bipartisan votes and awaiting Governor Pritzker's promised signature, would be the first state law to require annual independent audits of frontier developers' safety frameworks, alongside incident reporting on a 24-to-72-hour clock. The drafting is deliberately narrow — catastrophic risk only, developers above 500 million dollars in revenue — which Akerman reads as a play to become the de facto national baseline rather than a preemption target. Given what just happened to Colorado's broader approach, that calculation looks shrewd.