AI-Assisted Inventions: Human Contribution Is Central to Inventorship

On June 9, 2026, the Korean Intellectual Property Office announced guidance on proper patent filing in the age of artificial intelligence. The core message is clear: even when AI is used in the inventive process, a human must make a substantial creative contribution to be recognized as the legitimate inventor.

For companies and researchers using AI as an R&D tool, this is not merely policy news. It is a practical filing checklist that should be reviewed before preparing a patent application.

The key issue is human contribution, not the AI tool

Under patent law, the right to obtain a patent belongs to the inventor or the inventor's successor. AI itself cannot be named as the inventor. A person must contribute to the creation of the technical idea.

If a user gives a general prompt to generative AI and files the output as-is, registration may be difficult, and even a granted patent may face invalidity issues. The application should show who defined the technical problem, who selected the useful result from AI-generated candidates, and who modified the structure, process, algorithm, or experimental conditions.

Keep R&D notes and inventor contribution records

If inventorship is questioned during examination, the examiner may request materials showing human contribution, such as R&D notes or inventor statements. For AI-assisted inventions, companies should keep records before drafting the specification.

• Records showing that a human defined the technical problem and solution direction
• AI prompts, input data, model used, and date of use
• Human selection, modification, combination, and verification of AI outputs
• Experimental design, simulation conditions, and repeated verification results
• Specific contributions of each co-inventor

Where joint inventorship may be an issue, it is better to build records during the project rather than reconstruct them immediately before filing.

AI-generated test data must be verified

One of the most important practical warnings concerns AI hallucination. AI may generate nonexistent technical facts, unsupported effects, or unverified test results. Such content should not be presented in a specification or office-action response as if it were real experimental data.

The risk is especially high in pharmaceuticals, advanced materials, batteries, biotechnology, and chemistry, where effect data can strongly influence patentability. If AI suggests candidate compounds or performance values, those results should be experimentally verified before being used as filing materials.

Different AI invention types require different analysis

1. Inventions relating to AI itself: An abstract algorithm is not enough. The application should describe a concrete technical means implemented through information processing or hardware.
2. Inventions that include AI as a component: It is not enough to say that AI replaces human work. The application should explain the AI processing structure, training data design, inference flow, control logic, and technical effect.
3. Inventions derived using AI as a tool: The inventor should verify the AI result and convert it into a reproducible technical disclosure satisfying specification requirements.

Confidential information entered into AI tools must be controlled

Before using an AI tool, companies should check whether input data may be used for external model training or service improvement. Unpublished invention details, experimental data, customer materials, source code, manufacturing conditions, and composition ratios may affect novelty and trade secret protection.

Enterprise AI settings, access control, data retention policies, and restrictions on external tools should be reviewed before sensitive information is entered.

Pine IP Firm's view

AI can accelerate invention development, but it also raises inventorship, data verification, specification reliability, and trade secret issues. A strong AI-assisted patent application should show the human technical judgment behind the AI output. Filing the AI draft as-is is risky; using AI as a documented R&D support tool can improve both speed and quality.