How researchers took over ChatGPT accounts via artifacts exploit

Researchers at Parameter Labs have published a report, titled “Do the Math: How We Took Over ChatGPT Accounts,” on the company’s site under the URL parameter.ai/labs/we-found-a-gap-at-openai. The report was circulated on Hacker News and picked up for notice on October 31, 2026.

The title indicates a claim of having gained unauthorized access to ChatGPT accounts, attributed to an exploit involving what the headline describes as “artifacts.” Beyond the title and URL, no further text of the report has been made available to this desk for review. The Specola has not seen the body of the report and cannot characterize its findings, methodology, or scope.

OpenAI, the maker of ChatGPT, has not been quoted or cited in the material supplied. No statement from OpenAI accompanies this notice.

The matter was surfaced to this observatory through secondary notice rather than direct review of primary source text. Readers should treat the claim in the title as exactly that, a claim, pending fuller documentation.

This bulletin records only what the available material supports: the existence of a published report under the stated title and address, its circulation on Hacker News, and the date of notice. No additional detail of the alleged exploit, its mechanism, or its confirmation by any party can presently be reported.