Everyday AI data handling
Keep new ideas private during AI research
Use public background and general questions for initial AI-assisted research, and keep confidential invention detail out until the authorised legal/IP owner reviews the planned disclosure. Record the information, service, recipients and purpose. The legal effect of sharing depends on the facts and jurisdiction; a private account or no-training statement does not decide patent or confidentiality rights.
For R&D security and patent counsel
Give counsel a precise proposed disclosure, not a vague AI label.
Fictional research scenario and marker only. No actual invention, patent search or legal permission is supplied.01
Separate public research from the invention itself
A researcher can ask about established terminology or public technical background without describing the distinctive confidential mechanism being developed. Identify what is already public and what derives from the organisation’s unpublished work before preparing the prompt.
The exercise uses a generic question about folding supports and a separate marker standing for unpublished detail. It contains no actual invention. Its purpose is to demonstrate how a useful background question can remain separate from a pre-filing disclosure decision.
02
Describe the proposed recipients and information
Record the exact material that would enter the service, including diagrams, experimental values, claim drafts or automatically attached context. Identify the account, model/provider route and any additional connected service. A broad description such as “research with AI” is not enough for a meaningful review.
Retain the distinction between an input to a provider and a public publication. Do not assume either a particular legal consequence or complete confidentiality from the interface alone. Give the legal/IP owner the facts and applicable service terms needed to assess the intended action.
04
Use the worked fictional boundary
In the populated example, the team wants a general explanation of terminology for folding supports. The research prompt contains only that broad background question. A separate unpublished-feature marker is held out of the prompt until the fictional IP owner completes a disclosure review.
The outcome is intentionally narrow: proceed only with the generic synthetic background exercise, not the withheld invention detail. If the research later needs a distinctive mechanism or data, reopen the review rather than quietly adding it to a follow-up prompt.
| Example input | Classification in the exercise | Decision |
|---|---|---|
| General folding-support terminology | Public-style background question | Use only the generic exercise |
| SYNTHETIC_U09_UNPUBLISHED_DETAIL | Stands for confidential invention information | Withhold pending IP review |
| A later mechanism-specific follow-up | New proposed disclosure | Reopen the decision |
05
Record the decision and scope changes
Keep the information categories, recipients, purpose, reviewer and decision reference together. If a provider, account, attachment or research purpose changes, reassess the affected scope. Review generated output before treating it as reliable research or sharing it further.
No AI search, patent search or legal analysis was performed by the downloadable files. They contain only a fictional case and a clean prompt. Supported endpoint data controls can be part of an authorised workflow, but they do not replace the legal/IP owner’s disclosure decision.
Put it into practice
Public-background and invention-detail exercise
Separate a generic research question from confidential detail and record the authorised review needed for a change.
Fictional research scenario and marker only. No actual invention, patent search or legal permission is supplied.
| Material | Worked example | Outcome |
|---|---|---|
| Public-style background | General terminology for folding supports | Generic exercise only |
| Unpublished detail | Synthetic feature marker | Withheld |
| Recipient/purpose change | New service or mechanism-specific question | New review required |
Work through your review
Use the checks to organise the evidence you need. Your selections stay in this tab.
0 of 3 reviewed
Example files for this task
Keep the source material and the instructions together. You can also download the complete worksheet or matrix as CSV.
README.mdInspect
# Pre-disclosure research exercise
All material is fictional. This is not legal advice, a patentability assessment or permission to disclose an invention. No AI or patent database is queried.
Use public-background-query.txt to see the generic question. The marker in withheld-detail.txt stands for information that remains outside it. fictional-decision.md explains the worked outcome. For real work, use the responsible legal/IP owner and record the exact material, recipients and purpose.
## Guide and sources
Canonical guide: https://aona.ai/resources/guides/unpublished-inventions-ai-patent-research/
Source review: 2026-09-21
- USPTO: Guidance on use of AI-based tools: https://www.govinfo.gov/content/pkg/FR-2024-04-11/html/2024-07629.htm
public-background-query.txtInspect
SYNTHETIC BACKGROUND RESEARCH QUESTION
Explain common terminology used for folding supports in publicly available technical descriptions.
This general exercise includes no unpublished mechanism, experimental values or claim draft.
No research query has been submitted.
withheld-detail.txtInspect
SYNTHETIC_U09_UNPUBLISHED_DETAIL
This marker stands for confidential invention detail in the exercise.
No actual invention is described. Keep this separate from the background prompt.
fictional-decision.mdInspect
# Worked fictional disclosure boundary
Purpose: understand general terminology for folding supports.
Permitted exercise input: public-background-query.txt only.
Withheld category: unpublished mechanism/detail represented by SYNTHETIC_U09_UNPUBLISHED_DETAIL.
Reviewer in the exercise: SYNTHETIC_IP_OWNER.
Illustrative outcome: proceed only with the generic background exercise. No disclosure of the withheld category is authorised.
Recheck: mechanism-specific follow-up, attachment, provider/account or recipient changes.
This grants no real permission and makes no novelty or patentability finding.
disclosure-review.csvInspect
field,actual_record,owner,status
Information and public/confidential basis,TO_RECORD,ASSIGN,UNREVIEWED
Attachments and automatic context,TO_RECORD,ASSIGN,UNREVIEWED
Service account and recipients,TO_RECORD,ASSIGN,UNREVIEWED
Purpose and jurisdiction-specific advice,TO_RECORD,ASSIGN,UNREVIEWED
Authorised decision reference,NOT OBTAINED,ASSIGN,OPEN
Before you proceed
Keep these distinctions clear
- Removing the company name but keeping the mechanism
- Distinctive technical substance may remain confidential even when obvious identifiers are absent.
- Treating a service setting as a patent opinion
- Record the actual disclosure facts and obtain the appropriate legal/IP decision.
Apply it to employee AI use
Bring your actual data path.
Aona can help evaluate supported employee prompt/file policies for an already authorised workflow.
Aona does not assess patentability, interpret filing requirements or guarantee that a disclosure preserves rights.
Use a permitted synthetic background example to scope the input control, keeping invention-disclosure authority with the legal/IP owner.
Review your use caseFAQ
Questions for this decision
Does sending an invention to an AI service automatically destroy patent rights?
Does a private account or no-training promise supply permission?
Can initial research use only public background?
Does the pack contain a real invention or completed prior-art search?
Evidence behind the guide
Sources and scope
Prepared by Aona. Sources checked 2026-09-21. The cited material supports the specific points below; it does not certify a product or your use case.
- USPTO: Guidance on use of AI-based tools
Nonbinding USPTO-practice guidance discussing confidential information, AI tools and cross-border concerns.
regulator · checked 2026-09-21