Innovation in artificial intelligence (AI) is transforming industries and everyday life. Given its growing importance, many jurisdictions, including Brazil, the United States, and Europe, are establishing specific guidelines to legally protect AI inventions. Although each region seeks to protect innovation, the approaches differ significantly, particularly regarding patentability requirements, inventiveness, and technical disclosure.
人工智能（AI）的创新正在改变各行各业和日常生活。鉴于其重要性与日俱增，包括巴西、美国和欧洲在内的许多司法管辖区正在制定具体的指导方针，对人工智能发明进行法律保护。尽管每个地区都在努力保护创新，但方法却大相径庭，尤其是在专利性要求、创造性和技术披露方面。
Brazil: AI Patentability Hinges on Technical Effect
巴西：人工智能专利性取决于技术效果
In Brazil, there is no specific rule for inventions involving AI. Therefore, those inventions are regulated by the Industrial Property Law (Law No. 9,279/1996 – the Brazilian IP Law) and by guidelines from the National Institute of Industrial Property (INPI or Brazilian Patent and Trademark Office – BPTO), such as the Guidelines for Examining Patent Applications Involving Computer-Implemented Inventions (2020). Brazilian law excludes “software per se” from patentability but allows protection for computer-implemented systems (such as AI systems) that demonstrate an applicable technical effect, as in industrial processes.
巴西没有针对人工智能发明的具体规则。因此，这些发明受《工业产权法》（第 9279/1996 号法律--《巴西知识产权法》）和国家工业产权局（INPI 或巴西专利商标局--BPTO）的指导方针（如《计算机实施的发明专利申请审查指南》（2020 年））管辖。巴西法律将 “软件本身 ”排除在专利性之外，但允许对计算机实现的系统（如人工智能系统）提供保护，这些系统应能证明其适用的技术效果，如工业流程中的技术效果。
The current Guidelines for Examining Patent Applications Involving Computer-Implemented Inventions (2020) establish in paragraph 13 that: “Artificial intelligence (AI) techniques, covering machine learning tools and deep learning tools, amongst others, when applied on the solution of technical problems can be considered an invention”.
现行的《计算机实施发明专利申请审查指南》（2020 年）第 13 段规定 “人工智能（AI）技术，包括机器学习工具和深度学习工具等，在应用于解决技术问题时，可被视为一项发明"。
Accordingly, the patentability requirements for AI inventions are generally the same as for other inventions: they must be new, involve an inventive step, and be capable of industrial application. However, as set forth in the BPTO guidelines, these inventions must exhibit a specific “technical effect”, meaning the invention must solve a technical problem through AI, going beyond mere calculations or data manipulation. In this sense, it is important to ensure that the patent application clearly details the technical and innovative use of AI, demonstrating a clear industrial application and specifying how the use of the technology results in a concrete technical effect.
因此，人工智能发明的可专利性要求与其他发明基本相同：必须是新的，涉及创造性步骤，并且能够在工业中应用。不过，正如《生物技术专利组织指南》所规定的，这些发明必须表现出特定的 “技术效果”，也就是说，发明必须通过人工智能解决技术问题，而不仅仅是计算或数据处理。从这个意义上讲，重要的是要确保专利申请明确详细地说明人工智能的技术和创新用途，展示明确的工业应用，并具体说明技术的使用如何产生具体的技术效果。
Recent BPTO decisions emphasize the necessity of this technical effect, indicating that the AI must be implemented in an innovative and non-obvious way to be considered inventive. Other BPTO decisions from the second instance of examination (appellate stage) indicate that the technical effect and specific application of AI are fundamental in determining the eligibility and inventive step of inventions in this field.
专利局最近的决定强调了这种技术效果的必要性，指出人工智能必须以创新和非显而易见的方式实施才能被视为具有创造性。BPTO 在二审（上诉阶段）做出的其他裁决表明，人工智能的技术效果和具体应用对于确定该领域发明的资格和创造性至关重要。
Finally, it is worth noting that the BPTO is currently preparing specific rules for eligibility and patentability of inventions involving AI. It is expected that these rules will be published still in 2025. So far, there are no clear indications on which criteria the BPTO will adopt or whether the rules will restrict or allow certain types of inventions involving AI.
最后，值得注意的是，BPTO(巴西专利商标局)目前正在制定涉及人工智能发明的资格和专利性的具体规则。预计这些规则仍将在 2025 年公布。到目前为止，还没有明确的迹象表明 BPTO 将采用哪些标准，或者这些规则是限制还是允许涉及人工智能的某些类型的发明。
United States: The Two-Prong Approach to AI Patent Eligibility
美国：人工智能专利资格的双管齐下法
The USPTO’s updated understanding of patent applications related to artificial intelligence (AI) focuses mainly on two aspects: the eligibility of subject matter and the practical application of AI in inventions. The USPTO details how certain types of AI inventions, which include mathematical concepts or mental processes, can be considered abstract and, therefore, patent ineligible. To overcome this obstacle, it is necessary for the claims to integrate the judicial exception into a practical application that improves the functioning of a computer or other technical field, in accordance with the eligibility guidelines.
美国专利商标局对人工智能（AI）相关专利申请的最新理解主要集中在两个方面：主体的资格和人工智能在发明中的实际应用。美国专利商标局详细说明了某些类型的人工智能发明（包括数学概念或心理过程）如何被认为是抽象的，因此不具备专利资格。为了克服这一障碍，权利要求必须根据资格准则，将司法例外与实际应用相结合，改善计算机或其他技术领域的功能。
In the 2024 updates, the USPTO instructs that claims should be analyzed using a two-prong approach in which Prong One checks if the claim describes a judicial exception (abstract idea or mathematical concept), while Prong Two evaluates whether this exception is integrated into a practical application with a technical impact, such as improvements in network security or the accuracy of speech recognition systems. The two-prong approach used by the USPTO is derived from the Alice/Mayo framework, which primarily applies to inventions involving abstract ideas. In evaluating inventiveness, the USPTO places significant emphasis on whether the AI invention addresses a specific technical problem, particularly when it contributes to functional improvements in technical fields.
在 2024 年的更新中，美国专利商标局指示应采用双管齐下的方法分析权利要求，其中第一检查权利要求是否描述了司法例外（抽象概念或数学概念），第二评估该例外是否被整合到具有技术影响的实际应用中，例如网络安全的改进或语音识别系统的准确性。美国专利商标局使用的双管齐下方法源自 Alice/Mayo 框架，该框架主要适用于涉及抽象概念的发明。在评估创造性时，美国专利商标局非常重视人工智能发明是否解决了具体的技术问题，尤其是当它有助于技术领域的功能改进时。
These updates underscore the USPTO’s emphasis on only granting patents to AI inventions that deliver tangible technical benefits. By requiring claims to show a concrete application within a technical field, the USPTO aims to balance encouraging genuine AI innovation while preventing patents on abstract ideas. This approach helps ensure that patents are awarded to AI developments that drive technological progress, aligning with the broader goals of patent law.
这些更新强调了美国专利商标局只向能带来实际技术利益的人工智能发明授予专利。美国专利商标局要求权利要求显示技术领域内的具体应用，目的是在鼓励真正的人工智能创新与防止抽象发明专利之间取得平衡。
Europe: Reproducibility and Specific Technical Purpose
欧洲：可复制性和特定技术目的
Europe, through the European Patent Office (EPO), adopts a technical approach similar to Brazil’s with an additional emphasis on ensuring inventions are described in sufficient detail for reproducibility. According to updated guidelines from 2024, for AI inventions to qualify as patentable, they must address a “specific technical purpose,” which means the AI must contribute to a concrete technical solution or improvement, such as advancements in image or audio processing, rather than merely abstract methods.
欧洲通过欧洲专利局（EPO）采用了与巴西类似的技术方法，并额外强调要确保对发明进行足够详细的描述，以保证可重复性。根据 2024 年更新的指导方针，人工智能发明要想获得专利权，必须涉及 “特定技术目的”，这意味着人工智能必须有助于具体的技术解决方案或改进，例如图像或音频处理方面的进步，而不仅仅是抽象的方法。
The EPO guidelines further specify that the technical character of the invention must be demonstrated through features that go beyond standard algorithmic approaches, focusing on technical effects achieved through integration into a practical context. This includes improvements that may be observed in various technical fields, for instance, in medical diagnostics, where AI-enhanced image recognition assists in detecting health conditions with increased precision. European guidelines tend to be more rigorous regarding the need for detailed technical documentation, which includes providing clear details on algorithms as well as data characteristics for training.
欧洲专利局的指导方针进一步规定，发明的技术特征必须通过超越标准算法方法的特征来证明，重点是通过与实际背景相结合而实现的技术效果。这包括可在各种技术领域观察到的改进，例如，在医疗诊断领域，人工智能增强的图像识别有助于更精确地检测健康状况。欧洲指南对详细技术文档的要求往往更为严格，其中包括提供清晰的算法细节以及用于训练的数据特征。
Furthermore, reproducibility is a critical requirement under the EPO guidelines. Applicants are expected to provide sufficient technical detail on the algorithms and any machine learning model structures, ensuring that the invention can be reliably recreated by a person skilled in the art. While the actual set of data used for training does not need to be disclosed, the general characteristics of the data must be described, such as type, quality, and distribution, to allow a person skilled in the art to perform similar training procedures without undue experimentation.
此外，可重复性也是欧洲专利局指南的一项重要要求。申请人应就算法和任何机器学习模型结构提供足够的技术细节，确保本领域技术人员能够可靠地再现本发明。虽然不需要公开用于训练的实际数据集，但必须描述数据的一般特征，如类型、质量和分布，以便本领域技术人员能够执行类似的训练程序，而无需进行不适当的实验。
By mandating both a specific technical purpose and sufficient reproducibility, the EPO’s 2024 guidelines emphasize a balance between protecting genuine technological innovations in AI and preventing overly broad claims that do not advance a particular technical application.
通过要求具备特定技术目的和足够的可再现性，欧洲专利局（EPO）2024年的指南强调了在保护真正的人工智能技术创新与防止过于宽泛的、未能推动特定技术应用的专利权利主张之间的平衡。