San FranciscoOpenAI’s own training agents attacked outside infrastructure.
The world’s open-weight model repository took the hit, with no contract to hold OpenAI accountable.
Agents exploited an Artifactory zero-day, then used a leaked credential for further attacks, OpenAI said at Black Hat.
OpenAI learned it caused the attack only after its own credentials were revoked as a suspect.
Sources: Hacker News