128 THE US-ISRAEL | Legal Review 2025/26 United States: Obviousness and Hindsight US obviousness analysis under §103 is framed through Graham factors and often litigated with hindsight concerns. AI introduces new hindsight pathways: if an AI can generate a solution instantly today, decision-makers may unconsciously assume it was easy then. Practitioners should therefore document why the path was non-trivial at the relevant priority date—what failures occurred, what constraints mattered, and why the solution was not apparent without the inventor’s insight. Israel: Obviousness Despite the obvious differences between Israeli law and US or EPC law, inventive step is assessed in very similar ways across all three jurisdictions. Furthermore, because most patent applications filed in Israel are also filed in the US and the EPO, the conclusions reached in those jurisdictions have an impact on how an Israeli patent application is perceived. Enablement and Sufficiency: Fluent Text Is Not Adequate Disclosure AI-assisted drafting creates a different class of risk: applications that read well but do not teach. Under Section 12 of the Israeli law, a patent specification must contain sufficient information to allow a person skilled in the art to carry out the invention. Under 35 U.S.C. §112(a), the specification must enable the full scope of the claims and provide adequate written description. Under Article 83 EPC, the application must disclose the invention in a manner sufficiently clear and complete for it to be carried out by the skilled person. AI can produce plausible descriptions, but plausibility is not enablement. A well-written embodiment that omits critical parameters, experimental conditions, or algorithmic details may fail under Section 12, §112, or Art. 83. AI is also prone to generating “generic” descriptions that overreach, which can later invite enablement challenges when claims are broad. For practitioners, the lesson is straightforward: AI may help with language, but it cannot replace technical verification. Every AI-generated technical statement should be treated as untrusted until confirmed. If AI suggests an embodiment the inventor never built or tested, the practitioner must decide whether the disclosure is sufficiently supported or whether inclusion increases risk. Inventorship Versus AI Assistance: Doctrinal Fault Lines Inventorship is not a formality; it is a validity and ownership gateway. AI-assisted ideation pressures the doctrine in all three jurisdictions, albeit differently. United States: Conception, Joint Inventorship, and “Recognition” Problems In the United States, conception is “the formation in the mind of the inventor of a definite and permanent idea of the complete and operative invention.” Federal Circuit decisions emphasize that an inventor must contribute to conception of at least one claim element; mere reduction to practice, routine experimentation, or assistance does not suffice. Pannu v. Iolab Corp. highlights that incorrect inventorship can render a patent invalid, though correction may be possible under certain circumstances. AI complicates this analysis when the human’s role is primarily prompt design and evaluation. If a user asks an AI system for solutions and the AI proposes the key inventive feature, a later challenger may argue that the human did not conceive the feature—only recognized it. Recognition may not be enough if the human cannot show a mental conception independent of the AI output. In litigation, the distinction could become central: did the inventor have the definite and permanent idea before interacting with AI, or only after the AI proposed it? The problem is not theoretical. In inventorship disputes, courts examine contemporaneous evidence: lab notebooks, emails, drafts, and declarations. If the record shows that the “first appearance” of the key feature was an AI output, a challenger may frame the human as a selector rather than a conceiver. While courts have not yet articulated a stable rule for “AI selection,” practitioners should not assume selection is safe.
RkJQdWJsaXNoZXIy MjgzNzA=