# Using AI to draft an invention disclosure: what to check before you submit

URL: https://zernyxhub.com/blog/ai-invention-disclosure
Author: Sahil Sheikh, Regd. Patent & Trademark Agent, cRTTP
Published: 2026-11-01
Category: Technology transfer

> AI invention disclosure checklist for Indian faculty: confidentiality, inventorship under Section 6, invented data and citations, and what the law says on AI.

**Short answer:** AI can help structure an invention disclosure, but the inventor must check five things before submitting: that no confidential detail went into an unapproved tool, that only humans who conceived the invention are named as inventors, that every data point and citation is real, that the novelty claims are accurate, and that the best method is fully described.

Many researchers now open an AI chat window before they open the disclosure form. It is easy to see why: a disclosure asks for a structured account of the problem, the solution, the data and the prior art, and AI tools are good at structure. The risk is that they are also good at sounding right while being wrong, and that whatever you paste in may leave your control.

This checklist is for faculty, research scholars and postdocs at Indian institutions. It covers what to check before you submit an AI-assisted disclosure to your technology transfer office or IPR cell. For the disclosure itself, start with our guide to [writing an invention disclosure](https://zernyxhub.com/blog/invention-disclosure).

## What can AI do well in a disclosure, and what can it not?

| Task | AI is useful for | You must still do |
|---|---|---|
| Structure | Turning lab notes into the form's headings | Confirm each heading says what your data shows |
| Plain-language summary | A first draft of the problem and solution | Remove claims the data does not support |
| Prior art | Suggesting search terms | Run the search yourself and read the documents |
| Alternatives and variants | Listing possible embodiments | Keep only those you could actually make |
| Data | Nothing safely | Copy figures from your records, not from the draft |
| Inventorship | Nothing | Decide with your co-workers who conceived what |

The pattern is simple. Use AI for shape and language. Never let it be the source of a fact.

## Check 1: did anything confidential go where it should not?

An invention disclosure is confidential until the institution decides to file or to release it. Patent rights depend on that. Under the [Patents Act, 1970](https://ipindia.gov.in/frontend/pdf/patents/1_113_1_The_Patents_Act__1970___incorporating_all_amendments_till_1-08-2024.pdf), a prior publication can anticipate your invention, and Section 29 protects you only in narrow cases, such as where matter obtained from you was published without your consent and you filed as soon as reasonably practicable afterwards. India's grace period under Section 31 is narrower still: it covers display at a notified exhibition and a paper read before a learned society. Our post on [what happens when work is published before filing](https://zernyxhub.com/blog/published-before-filing) explains the consequences.

Whether typing unpublished details into an AI tool counts as making them public has not been tested by an Indian court or the Patent Office. That uncertainty is the point. Until it is settled, act as though it could matter.

1. Read your institution's policy on AI tools and research data before you start.
2. Use only a tool your institution has approved for confidential work, under terms that say your input is not used for training or shared.
3. If no such tool exists, use AI only on a non-confidential outline: headings, generic wording, the form's structure. Add the technical detail yourself, offline.
4. Never paste sequences, structures, formulations, code or raw data into a public chatbot.
5. If the work includes patient or participant data, your institution's data protection and ethics rules apply as well.

## Check 2: are the inventors the right people?

AI does not change who the inventor is. Section 6 of the Patents Act lets an application be made by the "true and first inventor", that inventor's assignee, or the legal representative of a deceased inventor. Section 2(1)(y) defines the true and first inventor by excluding the first importer of an invention into India and a person to whom it was first communicated from outside India. Both provisions are written for people.

On the question of AI as an inventor, an EU IP Helpdesk [report dated 27 July 2026](https://intellectual-property-helpdesk.ec.europa.eu/news-events/news/indias-patent-office-rejects-ai-inventor-landmark-dabus-decision-2026-07-27_en) says the Indian Patent Office refused the DABUS application, which named an AI system as inventor. We have not seen the Controller's order itself, so treat the details as reported.

If your institution may file in the US, note the USPTO's position. Its [revised inventorship guidance for AI-assisted inventions](https://www.federalregister.gov/documents/2025/11/28/2025-21457/revised-inventorship-guidance-for-ai-assisted-inventions), published on 28 November 2025, replaced the February 2024 guidance. It says only natural persons can be inventors, treats AI systems as tools, and asks whether a person conceived the invention. Where several people are involved, each must have contributed significantly to conception or reduction to practice.

**Who goes on the disclosure as an inventor**
Inventor:
- Conceived the solution or a claimed part of it
- Directed how the AI was used to reach it
- Can explain why it works
Not an inventor:
- The AI tool, however much it wrote
- Someone who only ran the tool or the assay
- A supervisor who only funded the work
AI is a tool, like lab equipment. Inventorship follows human conception.

For the rules on students, co-authors and supervisors, see [who is an inventor](https://zernyxhub.com/blog/inventorship-students-coauthors).

## Check 3: is every fact in the draft real?

AI drafts fail in predictable ways. Go through the draft line by line with your lab notebook open.

| What to check | Typical AI error | How to check it |
|---|---|---|
| Numbers | Rounded, swapped units, invented precision | Match each figure to a dated record |
| Citations | Papers or patents that do not exist | Open every reference; delete any you cannot find |
| Prior art claims | "No prior work exists" with no search | Search IP India, WIPO and Espacenet yourself |
| Scope | Broader than your data supports | Cut every embodiment you have not tested or cannot justify |
| Mechanism | A plausible but unproven explanation | Say "we believe" or remove it |

The prior art row matters most. A disclosure that overstates novelty wastes the TTO's time and money. Our [prior art search guide for faculty](https://zernyxhub.com/blog/prior-art-search-faculty) shows how to run a basic search in an afternoon.

## Check 4: would the disclosure support a complete specification?

The TTO will draft a patent application from your disclosure. Under Section 10(4), a complete specification must fully and particularly describe the invention and how it is performed, and disclose the best method of performing it known to the applicant. AI drafts tend to stay general. Make sure yours includes the working conditions, concentrations, parameters and the variant that actually works best, because a gap here becomes a weakness in the patent later. Our post on [turning a paper into a patent application](https://zernyxhub.com/blog/paper-to-patent-application) explains what a drafter needs from you.

## Check 5: what records should you keep?

Keep a short note with your disclosure: which tool you used, for which sections, on what date, and what you changed. If inventorship or confidentiality is ever questioned, that note answers it. Under the 2024 amendments to the Patents Rules, the Patent Office also issues a [certificate of inventorship](https://www.pib.gov.in/PressReleasePage.aspx?PRID=2015234&reg=48&lang=2), so accuracy about who invented what now shows on a public document.

**Before you press submit**
1. Confidentiality: Approved tool, or no confidential input
2. Inventors: Humans who conceived it, agreed in writing
3. Facts: Every number and citation traced
4. Best method: Working parameters included
5. Record: Tool, sections, date, changes

## What should you tell your TTO?

Be open about AI use. A short note at the end of the disclosure is enough, and it helps the office in three ways.

First, the patent drafter will know which passages to check against the data before relying on them. Second, the office can judge any confidentiality risk early, while there is still time to file before something becomes public. Third, if a question about inventorship or originality arises during examination or in a licence negotiation, the record already exists.

A useful note covers four points:

1. The tool used, and whether it was an institution-approved tool.
2. Which sections it helped draft, for example the background and the summary.
3. What you entered: an outline only, or technical detail.
4. Confirmation that all data, figures and references were checked by you against your records.

Some offices may not yet have a policy on AI-assisted disclosures. If yours does not, the note still protects you, and it gives the office a model to build one from.

## Common mistakes

- Pasting the full draft paper into a public chatbot "just to tidy the English".
- Keeping a reference because it looks right, without opening it.
- Letting the AI's broad description replace the narrow one your data supports.
- Listing a colleague as inventor because they ran the AI tool.
- Submitting a polished disclosure that says nothing about how the invention is actually made.

AI tools are worth using for disclosures, and refusing them entirely leaves faculty doing slow work by hand. The trade is that the checking time goes up even as the writing time comes down. Zernyx's TT Co-Pilot is an AI assistant for technology transfer, built for TTO teams and faculty on Zernyx's proprietary technology licensing dataset; whichever assistant you use, the five checks above stay with you.

## Common questions

### Can I use ChatGPT or another AI tool to write my invention disclosure?

Only with care. Check your institution's rules first, and use a tool the institution has approved for confidential work. Whether entering unpublished details into an AI tool could count as a public disclosure is untested in India, so treat the risk as real and keep sensitive detail out of public tools.

### Can an AI system be named as an inventor in India?

Section 6 of the Patents Act, 1970 lets the true and first inventor, their assignee or a deceased inventor's legal representative apply, and the law is read as referring to people. A July 2026 report says the Indian Patent Office refused the DABUS application naming an AI as inventor.

### Does using AI change who is an inventor?

No. The people who conceived the invention are the inventors; the AI is a tool, like lab equipment. The USPTO's November 2025 revised guidance takes the same position for US filings, which matters if your patent will be filed in the US.

### What do AI drafting tools most often get wrong in a disclosure?

They invent references, round or alter numbers, describe the invention more broadly than the data supports, and claim novelty without a search. Each statement in an AI draft needs checking against your lab records.

---

Published by Zernyx Hub. More: https://zernyxhub.com/llms.txt
