126 THE US-ISRAEL | Legal Review 2025/26 whether through explicit training, fine-tuning, retrieval augmentation, telemetry-driven updates, or other mechanisms. Even when user data is not retained verbatim, the system may abstract patterns. For patent practitioners, that distinction is crucial: an abstraction can still leak the inventive essence. The legal system’s “tool” framing, therefore, creates blind spots. Patent law presumes that the boundary between private and public is policed by human confidentiality obligations, contractual duties, and controllable dissemination channels. AI introduces an additional channel that is often opaque. Even if providers offer contractual assurances, the practical ability to prove what was stored, what was used for training, and what was later emitted is limited. Practitioners should treat this as a structural risk, not a mere technical detail. Anthropomorphism and the Illusion of Private Consultation The risk is amplified by psychology. LLMs converse fluently and contextually. Users tend to experience the interaction as consultation rather than computation. For patent professionals, this anthropomorphism can be hazardous because it encourages the sharing of half-formed concepts, problem–solution formulations, and strategic claim directions that would not be disclosed to an external party. The “friendly” nature of the interface can lead to careless disclosure. In professional-responsibility terms, this interacts with duties of confidentiality. Attorneys and agents owe duties to clients; in many jurisdictions, confidentiality is broader than privilege. If a practitioner inputs confidential invention details into an AI system without adequate safeguards, that act may be scrutinized later as a breach of duty, even if no harm is immediately apparent. A practitioner-warning perspective is therefore warranted: the reputational and ethical consequences may arrive before the doctrinal answers do. Confidentiality, Disclosure, and “Public Availability” The doctrinal core question is whether disclosure to AI can constitute making information “available to the public” (EPC Art. 54 and IL Section. 4) or otherwise constitute prior art (35 U.S.C. § 102). Neither system has yet produced a definitive, widely applicable precedent addressing AI chat disclosures. But patent law already contains tools for analyzing functional equivalents. Israel and Europe: “Made Available to the Public” Under Article 54 EPC or “Made Public” Under IL Section 4 Israeli law (Section 4) is pretty straightforward, stating that an invention is novel unless it was “made public.” Under the EPC, the state of the art includes everything made available to the public “by means of a written or oral description, by use, or in any other way.” The Boards of Appeal have long held that a disclosure is public if at least one member of the public could access it without confidentiality obligations and could understand it. The mechanism does not matter; what matters is accessibility and availability. If an AI system, after receiving a user’s confidential input, can provide substantively similar guidance to another user not bound by confidentiality, an opponent may argue that the invention’s teaching was made available to the public “in any other way.” Even if the output is not verbatim, the EPC recognizes implicit “AI can offer a “bear hug:” it feels supportive at first, but tightens until it compromises the very rights it was meant to help secure.”
RkJQdWJsaXNoZXIy MjgzNzA=