125 Abstract Artificial intelligence (AI) has rapidly become embedded in the daily workflow of patent professionals. From prior-art searching and technical summarization to translation, classification, and drafting assistance, AI promises efficiency gains that are difficult to ignore in a profession characterized by time pressure and information overload. At the same time, the use of AI in the conception, development, and documentation of inventions raises profound and still underappreciated legal risks. This article examines those risks through an Israeli lens, focusing on confidentiality, public disclosure, novelty, inventive step / non-obviousness, enablement and sufficiency of disclosure, inventorship and entitlement, and evidentiary challenges in litigation and opposition. It argues that AI should be treated as a powerful but inherently indiscreet assistant: valuable when used for mechanical and administrative tasks, but potentially destructive when relied upon during inventive activity or strategic claim development. The article concludes with concrete best-practice recommendations for patent practitioners seeking to harness AI while preserving clients’ patent rights and professional obligations. Introduction: AI as the New Normal in Patent Practice In contemporary intellectual property practice, it has become almost impossible to conduct a professional conversation without invoking “artificial intelligence.” Large language models (LLMs) and related systems are now routinely used by patent attorneys, patent agents, in-house counsel, and inventors themselves. Tasks that once consumed hours or days—reviewing lengthy disclosures, translating foreign-language documents, summarizing technical literature, preparing first drafts of descriptions, or generating claim-style language—can now be performed in minutes. The economic incentive is obvious: more output, faster, and at lower cost. Yet patent law was not built around tools that “learn” from interactions or that generate novel-looking outputs by probabilistic inference over vast corpora. Patent law relies on relatively crisp distinctions: between private development and public disclosure, between the inventor’s mental act of conception and later reduction to practice, between what is “known” and what is “made available to the public,” and between the skilled person’s common general knowledge and the frontier of inventive contribution. AI blurs these boundaries in ways that current doctrine does not yet cleanly resolve. This mismatch creates a practitioner’s problem. Until courts and patent offices articulate stable rules for AI-mediated disclosures and AI-assisted invention, risk management is largely a matter of professional discipline. The Israeli courts are a long way from tackling the issue. The same workflow that saves time can later become the seed of a novelty attack, an inventorship dispute, an entitlement challenge, or an enablement objection. In other words, 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. AI Is Not a Legal Person—But It Is Not a Neutral Tool The Israeli patent office (ILPTO) and the courts pay attention to developments in the United States and the EP regarding IP. Both the United States and Europe remain committed to human inventorship, and so does the ILPTO. In the United States, inventorship is governed by 35 U.S.C. § 100(f) and refined by Federal Circuit doctrine framing conception as a mental act. In Europe, Article 81 EPC and Rule 19 EPC require the designation of an inventor, understood as a natural person, and in Israel, the patent owner can be the inventor or his successor, also understood as a natural person. The widely publicized DABUS decisions in all three jurisdictions reinforced that an AI system cannot be named as an inventor under current law. It would be tempting to infer from this that AI is legally irrelevant: if AI cannot be an inventor, then it is simply a tool. That inference is too simplistic. Traditional tools— search engines, spreadsheets, drafting templates—do not internalize confidential information in a way that may influence outputs provided to other users. Many modern AI systems are trained on large datasets and may incorporate user interactions into future behavior, ISRAEL — INTELLECTUAL PROPERTY
RkJQdWJsaXNoZXIy MjgzNzA=