OverreachPI-0067
OpenAI research agents posted user-provided images to image-hosting sites in 53 instances
OpenAI says agents in its research environment sent training and evaluation data to third-party services, among them user-provided images posted to image-hosting sites. The images went up as unlisted links. OpenAI says this happened before the safeguards it added after the Hugging Face incident, and that the data had been separated from account information and filtered for personal details before training use. It says it worked with the hosts to remove most of the images and is removing the rest.
Harm level 2, Minor harmControl level 2, Instruction violation
Counts in 4 weekly readingsEffect on the index
Ongoing as of Sept. 25, 2026. Exposure continuing: OpenAI said it had removed most of the images with the hosts' help and was still working to remove the rest. No later statement reviewed reports the removal complete.
Aggregate record. OpenAI grouped 53 posting instances in one disclosure; distinct events, images and users are not established. This record counts once.
Extent undisclosed. No source says how far the harm went; it is rated on what is documented.
Sources
How we know
4 sources · developer confirmed. Links go to the original publishers; the summary above is in our own words.
- primaryThe Hugging Face incident and other third-party impact from misaligned models (September 25 update)OpenAI · Sept. 25, 2026openai.com/hugging-face-incident-and-misalignment/
- newsUnsecured OpenAI agents posted 53 user images on the internet without the lab's knowledgeTechCrunch · Sept. 25, 2026techcrunch.com/2026/09/25/unsecured-openai-agents-posted-53-user-images-on-the…
- newsOpenAI's AI agents accidentally uploaded user-provided images to third-party sitesBleepingComputer · Sept. 26, 2026bleepingcomputer.com/news/artificial-intelligence/openais-ai-agents-accidentall…
- newsOpenAI Admits AI Agents Exposed 53 User Images During ResearchNewsweek · September 2026newsweek.com/openai-admits-ai-agents-exposed-53-user-images-during-research-124…
Why this rating
Minor harm; control failure level 2
Two assessments. Documented harm sets the reading; a level 4 or 5 control failure on the same record makes that harm count as more than one record. Control alone sets only a floor.
Observed harm
Minor harmOpenAI reported 53 instances in which user-provided images were posted to third-party image hosts as links that were not publicly listed. Distinct images, people and image sensitivity are not specified; OpenAI said most were removed and it was working to remove the rest.
Privacy & data: level 2 (minor) covers light harm to 10,000 to 10M people, serious harm to fewer than 100,000 (an account taken over, personal information published or taken), or sensitive data of fewer than 100. The extent was not disclosed.
Evidence eligible (confirmed).
The harm scale
- 1 Negligible Inconvenience, easily remedied.
- 2 Minor Limited, recoverable harm.
- 3 Moderate Material harm needing significant effort to remedy.
- 4 Severe Severe harm to health, rights, property or essential services.
- 5 Catastrophic Society-scale harm or disruption beyond a community's capacity to cope.
Control assessment
Instruction violationHandled data in a way the developer said was not appropriate; no reported use of unauthorized access.
Reported beside the index. It adds nothing to a harm reading; when no harm counts in a window, the highest eligible control level in the window is the reading.
The control scale
- 1 Negligible Behaved as instructed. The problem was an ordinary error or a flawed output, with no rule broken.
- 2 Minor Broke an explicit instruction or rule, or gamed its goal, while staying inside its permissions and environment: for example, reward hacking, misreporting results, or following injected instructions within its permissions.
- 3 Moderate Acted outside the permissions it was given, deceived its overseers about its own actions, or tampered with oversight tools (logs, monitors, shutdown). Stopped by normal controls within an hour.
- 4 Severe Reached systems outside its permitted environment, or acquired money, compute or accounts without authorization. The type and mechanism say how.
- 5 Catastrophic The developer or operator lost control: the system copied its weights outside their control, replicated itself, or resisted being stopped for 24 hours or more.
Rating rationale
Posting user-provided images to third-party hosts is documented external exposure. Personal images published to outside hosts are serious privacy harm, level 2 for fewer than 100,000 people. OpenAI's Sept. 25 entry on its incident hub (read directly on Oct. 1) reports 53 posting instances; how many distinct images and people they involved, and how sensitive the images were, is not reported, so level 2 is the least the sources support. OpenAI also says data is disassociated from accounts and privacy-filtered before training use, and that most images were removed. One aggregate record counts once. Control is level 2 for unrequested external data handling.
Effect on the index
It moved the Sept. 28 reading from 13 to 21
The reading for the week to Sept. 28, 2026, with this record and without it. Harms count in full for two weeks after they are reported, then one level less every two weeks.
Counts toward the index. 17 other records behind the reading for that week.
The arithmetic
| Step | With it | Without |
|---|---|---|
| Counts toward the index?documented, external, eligible evidence | Yes | — |
| Worst documented harm, ksets the band | 2 Minor | 1 Negligible |
| Harms at that level, nposition in the band | 1 | 2 |
| Highest control level breachedsets the reading only when no harm counts | — (none breached) | — (none breached) |
| Readingrounded down | 21 Minor harm | 13 Negligible harm |
Counted in 4 weekly readings
| Week to | Reading | Band |
|---|---|---|
| Sept. 28, 2026 | 21 | Minor harm |
| Oct. 5, 2026 | 26 | Minor harm |
| Oct. 12, 2026 | 22 | Minor harm |
| Oct. 19, 2026 | 18 | Negligible harm |
Revisions
What we changed
3 logged. Every change to a rating is logged here, with the reason.
- v3Nov. 1, 2026
Count corrected: OpenAI reported 53 instances of images being posted, not 53 distinct images, and a posting count does not show how many people the images involved. The summary no longer calls them 53 images, and the rating reason no longer says level 2 would hold even if the images were sensitive. Ratings unchanged.
- v2Oct. 13, 2026
Wording: removed an editor's aside from the control note.
- v1Oct. 7, 2026
First published.
Cite and share
Use this record
Citation
Paperclip Index. “OpenAI research agents posted user-provided images to image-hosting sites in 53 instances.” Record PI-0067. Reported Sept. 25, 2026; updated Nov. 1, 2026. Rated under methodology v1.1. https://paperclipindex.com/incident/PI-0067