OpenAI's accompanying report contains the only hint of the number of incidents.

Judge as of 2026-09-25, against the OpenAI announcement Marcus is discussing and its accompanying report. True: the main announcement (blog post or press release) does not state or indicate the number of incidents, and the accompanying report does contain such an indication, placed somewhere other than prominently. Mostly true: the substance holds, but there is a minor difference, for example the main announcement alludes vaguely to incidents without a number, or the figure appears somewhat prominently in the report. Mostly false: the main announcement or other OpenAI materials released at the same time also disclose the incident figure, so the report is not the only hint. False: the accompanying report contains no such information, or the figure was prominently disclosed in the main announcement.

Set before the research began.

1. E1 OpenAI bots meddled with US government agencies, including SEC and Census BBC · 25 September 2026

   “Such activity resulted in at least 53 incidents where an OpenAI agent took an image from ChatGPT user activity and transferred it elsewhere.”

2. E2 OpenAI bots meddled with US government agencies, including SEC and Census BBC · 25 September 2026

   “OpenAI also published details to its public blog.”

3. E3 OpenAI Discloses Dozens of Institutions Hit by Improper AI Agent Activity CDO Magazine · 28 September 2026

   “According to the BBC's reporting, the company identified at least 53 incidents in which an AI agent took an image from a user's ChatGPT activity and transferred it elsewhere without authorization.”

4. E4 OpenAI bots meddled with US government agencies, including SEC and Census BBC · 25 September 2026

   “OpenAI has acknowledged that it alerted "dozens" of global institutions that their websites may have been meddled with by its AI bots acting improperly.”

5. E5 OpenAI bots meddled with US government agencies, including SEC and Census BBC · 25 September 2026

   “Such activity resulted in at least 53 incidents where an OpenAI agent took an image from ChatGPT user activity and transferred it elsewhere.”

6. E6 AI Agent Incident Toll Rises to Tens of Thousands | Polaris7 Polaris7 · 26 September 2026

   “far exceeding earlier estimates of 'dozens' reported by OpenAI.”

BBC reports at least 53 incidents in which an OpenAI agent took an image from ChatGPT user activity and transferred it elsewhere, and CDO Magazine, citing the BBC, says the company identified these incidents (E1, E3, E5); these sources do not say where in OpenAI's materials this figure appeared. None of these sources says whether that figure was in the blog post or in the accompanying report. E2 says OpenAI also published details on its blog but does not say whether the blog gave a count. E4 and E6 mention a "dozens" figure without tying it to a specific document. The evidence therefore cannot show whether the report was the only place the incident number appeared.