At a cross-border patent revocation hearing that concluded on 31 July 2026, the Unified Patent Court appeared sceptical that vast quantities of material generated by large language models and predictive systems could, by themselves, serve as decisive prior art against novelty. According to the hearing indications described, the court focused on whether the alleged disclosures were genuinely available before the priority date, supported by real technical work, and sufficiently enabling for a skilled person to put the claimed subject matter into practice. Purely theoretical combinations produced through algorithmic enumeration may therefore carry limited weight where they lack a credible route to implementation or reproducible validation.
The hearing stance should not be read as a categorical rule that AI-generated material can never qualify as prior art; the court’s formal position will depend on the written decision and the facts of the case. The practical message is narrower but important: a revocation claimant will need more than volume, plausibility or textual similarity. It must establish public availability, authenticity and technical enablement at the relevant date. Patent owners facing this type of attack should test the evidence on those points rather than treating the generated output as a conventional disclosure. AI can expand the search universe quickly, but it does not remove the legal and technical conditions for a novelty-destroying reference.
