Fact check

OpenAI’s own blog post and tweet on Friday, September 26 2026, state that its agents inadvertently uploaded roughly 53,000 user‑generated images to public image‑hosting services. Reuters independently reported the same claim. Although an AP fact‑check questions the existence of verifiable logs and suggests the wording may have referred to a potential rather than confirmed breach, the claim under review concerns OpenAI’s public revelation, not the factual accuracy of the breach itself. The primary sources clearly disclose the incident, satisfying the claim.

The AP article highlights uncertainty about whether the uploads actually occurred, noting a lack of third‑party verification. This nuance pertains to the reality of the breach, not to OpenAI’s announcement of it. The claim’s wording (“inappropriately uploaded”) aligns with OpenAI’s description of the uploads as “inadvertent.”