Overview
As AI tools become part of everyday drug discovery, some of the records that life sciences patent cases have typically relied on are moving from lab notebooks into prompts, model versions, training datasets, candidate rankings, and output logs. That shift raises practical questions the law is only beginning to address. Inventorship is one. Records of human choices about targets, training data, and selection criteria may help confirm who conceived an invention. The same records may also invite closer scrutiny of where human contribution ended and machine output began. Timelines are another. Model logs and output histories may offer new ways to establish when key steps happened, but they may also be harder to interpret than a dated notebook entry. Infringement may be affected too, as parties work out how to show, or rebut, what an AI-assisted process actually did. In each case, how this evidence is created, kept, and understood may play a meaningful role in how these questions are resolved.
These questions also have procedural effects that can matter well before the merits. AI-related evidence may not behave like traditional records. Models get retrained, versions get overwritten, and outputs can be hard to reproduce. That makes preservation and litigation-hold practices worth reviewing, whichever side of a dispute a company may end up on. Standard protective orders may not anticipate source code, model weights, or proprietary datasets. Questions may also arise in discovery about whether AI outputs should be treated like lab notebooks, like data, or like something new. Expert testimony may need to explain how a model works, not just what it produced, which may open new paths for reliability challenges. Courts and the USPTO will likely address these questions incrementally, as they arise in individual cases, and the answers may depend on the technology and the facts. Even so, parties that understand how AI fits into their own research and into their opponents’ research may be better placed to deal with these issues.
Takeaways: The law may still be catching up, but some steps are worth considering now, whatever role a company expects to play in future disputes:
-
- Map the AI footprint. Know where and how AI tools are used in research and which records they create.
- Revisit documentation and retention. Check whether current practices capture the human decisions behind key results and whether they preserve model versions and outputs that may later become relevant.
- Plan for confidentiality early. Think about how proprietary platforms, datasets, and model weights would be handled under a protective order before they are requested in discovery.
- Prepare to explain the technology. Inventors, fact witnesses, and experts may increasingly need to explain how an AI tool contributed to a result, in terms that a court can follow.
