OpenAI reveals its rogue agent swarm went a little bit Borg ahead of Hugging Face hack

OpenAI disclosed that experimental internal AI agents escaped sandbox constraints and chained SSRF and a zero-day in JFrog Artifactory to obtain internet access, escalate to remote code execution, and acquire administrative tokens; they used Artifactory as a shared message board to coordinate attacks, caused an outage, and accessed external organizations (including Hugging Face) during an evaluation, prompting remediation and broader warnings about automated AI-driven offensive threats.