Fictional example 1
A 30-person biotech lab
Grant-funded, publishes in journals, uses AI for literature summaries and first drafts.
| Area | Risk level | What we found |
| Checking AI output | High | AI-drafted summaries reach grant reports without a set check of their references. |
| Funder, journal and regulator rules | High | Applications are drafted with AI, and the funder's current AI rules haven't been reviewed. |
| Confidential data | Moderate | The data rule doesn't cover contractors, who use a free chatbot. |
| Disclosure and authorship | Moderate | AI use is disclosed, but not the same way each time. |
Fictional example 2
A 12-person weight-management clinic
A clinical practice that uses AI to draft patient letters and handouts.
| Area | Risk level | What we found |
| Patient information | Critical | Patient details go into a free AI tool whose vendor has signed no business associate agreement. |
| Confidential data | High | There is no written rule on what may go into AI tools. |
| Checking AI output | Moderate | AI-drafted handouts reach patients without a clinician's review. |
| AI tools in use | Moderate | Three tools in use, one on a personal account, and no list. |
Fictional example 3
An 80-person medical writing firm
Writes manuscripts and reports for pharmaceutical clients, with a written AI policy.
| Area | Risk level | What we found |
| Contracts and ownership | Moderate | Client contracts don't cover AI use, or who owns AI-drafted sections. |
| Disclosure and authorship | Low | A standard disclosure line is in place and used on every manuscript. |
Fictional example 4
An 18-person regulatory consultancy
Prepares regulatory submissions for small drug developers.
| Area | Risk level | What we found |
| Confidential data | High | Client documents have gone into a general chatbot on a personal account. |
| Funder, journal and regulator rules | Moderate | Some of the firm's AI use may fall under a regulator's draft guidance, and nobody has checked which. |
| Checking AI output | Moderate | Drafts are reviewed, but AI-supplied citations aren't checked. |
Fictional example 5
A 40-person research foundation
A grant-funded nonprofit whose staff also serve on review panels.
| Area | Risk level | What we found |
| Funder, journal and regulator rules | High | Proposals under review were summarized with a chatbot. |
| Confidential data | Moderate | The data rule leaves out material others share in confidence. |
| Training | Moderate | One team has had AI training; another hasn't. |
Where the guidance comes from
Each finding in an IEACOR report is compared with published rules and guidance. In these examples, that guidance comes from:
- National Institute of Standards and Technology (NIST)
- National Institutes of Health (NIH)
- National Science Foundation (NSF)
- U.S. Food and Drug Administration (FDA)
- U.S. Department of Health and Human Services (HHS)
- U.S. Copyright Office
- International Committee of Medical Journal Editors (ICMJE)
Named as sources only. IEACOR is not affiliated with or endorsed by any of these organizations.