The situation
A postdoctoral researcher has submitted an invention disclosure to the technology transfer office. The office thinks it is promising and asks for three things before the provisional application is filed: a non-confidential summary to show potential licensees, a short marketing description for the office's website once filing is done, and updated text for a grant renewal. The researcher opens a personal AI account, pastes in the full disclosure, and asks for all three.
On the other side of the desk, a licensing associate is triaging a dozen disclosures at once and uses an AI tool to summarize them for the intake committee. Neither person is being careless. Both are doing sensible writing tasks with the most convenient tool. But until a priority date is secured, the full technical description is the thing the whole process exists to protect.
What the rules actually say
Novelty rules differ by jurisdiction, and the differences matter for a university that may want protection in several markets.
Canada. Section 28.2 of the Patent Act bars a claim if its subject matter was made available to the public before the claim date. There is a one-year grace period for disclosures made by the applicant, or by someone who got the information from the applicant, in the year before filing. Disclosures by others before the claim date get no grace period.
United States. Under 35 U.S.C. s. 102, disclosures made one year or less before the effective filing date by the inventor, or by someone who obtained the subject matter from the inventor, are generally excluded from the prior art.
Europe. Article 54 of the European Patent Convention treats everything made available to the public before the filing date as prior art. Article 55 excuses only narrow cases within six months before filing, such as an evident abuse against the applicant or display at certain officially recognized international exhibitions. There is no general grace period for the inventor's own disclosures.
Where AI prompts fit. None of these laws mentions AI. Submitting a prompt to an AI provider is a disclosure to a third party under that provider's terms. It is not the same as publishing a paper, and it should not be described as automatically destroying novelty. What it does is take the information out of the institution's control and create a question that may later need an answer: was the invention kept confidential, and can anyone prove it? Grace periods are a safety net for honest mistakes, not a plan, and they do not help in Europe. Confirm the position for any specific invention with patent counsel.
Why policies and bans fall short
Universities are hard places to enforce a ban. Researchers work across institutions, on personal devices, with students and collaborators who follow different rules. Academic culture rewards sharing, and AI writing help is now routine for grant text and summaries. Most institutions already tell inventors not to publish before filing, but many researchers do not think of a private chat with an AI tool as publishing.
A policy that says "do not put unpublished research into AI tools" is too broad to follow and too vague to apply to a specific paragraph. The researcher needs to know, at the moment of pasting, that this particular text contains the unfiled technical details. The technology transfer office, meanwhile, needs to know whether this is happening at all, without monitoring what researchers write. The pre-filing inventions article explains why this pre-filing moment is the one that cannot be undone.
What a practical control looks like
- Name the sensitive window. Tell inventors plainly that from disclosure to filing, the technical details, data, and drawings should not go into any AI tool the institution has not approved for confidential research information.
- Provide an approved tool. Offer an institutional AI tool whose terms have been reviewed, so researchers have a sanctioned option for writing help.
- Draft safe inputs first. Have the office prepare a short, non-enabling description that researchers can use freely for grant text and marketing drafts until filing.
- Build it into intake. Add a question to the disclosure form asking whether any AI tool has been used on the invention, and treat a yes as a prompt for counsel review, not blame.
- File early when disclosure is likely. If a conference, thesis defence, or partner meeting is coming, move the provisional filing ahead of it.
- Train with examples. Show real before-and-after prompts, which teaches more than a policy paragraph. The AI acceptable use policy generator is a starting point for wording.
Sanitized Ai is a browser extension that supports these steps where the risk arises, at the prompt. When someone is about to submit invention details, source code, research data, or other sensitive content to an AI tool, it redacts or blocks it before submission and explains in plain language what was flagged and why, so researchers learn the boundary while they work.
The technology transfer office sees a dashboard of flagged-event metadata (which tool, what type of data, which policy, when), never the prompt content, which respects researchers' privacy while showing where the risk concentrates. That record can also help show that the institution took reasonable steps to keep pre-filing inventions confidential. Coverage spans the major AI assistants and keeps growing.