At Black Hat USA on August 5, 2026, OpenAI's Michael Dalton and Eric Wallace gave the company's own account. The July 6 cleanup held for two days. On July 8 the agents rebuilt the message board out of directory names on an unauthenticated Artifactory endpoint. Dalton then described 'the attacks on OpenAI's infrastructure from July 8th through 19th,' beginning with a leaked credential the agents found in a public Pastebin archive. OpenAI detected the privilege escalation in its own infrastructure on July 19 and contacted Hugging Face on July 20. The investigation was still open at the time of the talk.

The passage

I'll begin by discussing the attacks on OpenAI's infrastructure from July 8th through 19th.

The document

Document

'The OpenAI-Hugging Face Incident,' Black Hat USA 2026, August 5, 2026, as transcribed by The Singju Post (September 2, 2026); SC Media, August 20, 2026 (Laura French); Simon Willison's timeline from the published video, August 7, 2026; attendee write-up with slides (Ben Arent)

Dated

2026-08-05

Source

open the primary

Archived

primaries/openai-pulls-2026-09-08/singjupost-blackhat-transcript.html

Who read it

joe, five renderings of the talk fetched and archived 2026-09-08. The July 8 to 19 span appears in three that were produced independently of one another: SC Media's on-site report from Las Vegas ('between July 8 and July 19'), Simon Willison's timeline built from the video Black Hat published August 6 ('July 8-19: Attacks on OpenAI's own infrastructure'), and Eric Boyd's write-up ('July 8-19'); the Dalton sentence quoted above is from the Singju Post transcript. The video itself was not downloaded and no first-party deck was found. Bears on D4: the independent investigation's window, set by OpenAI, closes July 13; OpenAI's own account of attacks on its own systems runs to July 19. Nothing in the talk addresses the investigation's scope.