Source: Goodwin - JDSupra

Goodwin - JDSupra addresses the critical tension between the accelerating field of AI-assisted drug discovery and the established principles of patent law, particularly in the United States. AI dramatically speeds up the research process, helping to design novel molecules and predict therapeutic candidates, with the economic implications projected to reach $13.4 billion by 2035. However, the commercial viability of these discoveries depends on patent protection, which mandates a human inventor who performed "conception"—the formation of a definite idea of the complete invention. Existing US case law holds that AI systems cannot be inventors, yet it has not clarified what level of human contribution is sufficient when AI performs the generative work. This uncertainty poses a risk that AI breakthroughs could be left unpatentable, thus undermining the economic incentive for innovation.

The core issue lies in modern AI workflows. When a scientist sets constraints for an AI model to design a novel compound, and the model outputs a structure later validated, the question of who conceived the invention is unclear. Merely posing a research goal or validating an output is insufficient for inventorship, but current practices involve human teams in framing the problem, selecting data, tuning models, and interpreting rankings. The authors suggest that companies must strategically design workflows to preserve and document human involvement, focusing on capturing human choices, rationales, and iterative refinements that support the claim of conception.

To mitigate patent risk, companies are advised to rigorously document the scientific team’s input, including why particular constraints or data were chosen, and how prompts and parameters changed over successive runs. This documentation establishes human contribution and builds credibility with patent examiners. Furthermore, patents require disclosure that enables a skilled person to make and use the invention. For AI-generated compounds, this requires disclosing enough detail on inputs and scientific rationale to avoid relying on the "black box" explanation of "The model said so," which would fail the enablement test. The authors conclude that the patent litigation expected in the next two to three years will establish crucial precedents that will either sustain or fundamentally restructure pharmaceutical innovation in the age of AI.