OpenAI paused all training runs... ALIGNMENT FAILURE
From Wes Roth · · 15 min

The player loads from YouTube only when you click. It uses YouTube's privacy-enhanced mode.
If the video does not play, the owner may have turned off embedding. Watch it on YouTube instead.
Description
OpenAI’s alarm went off. The automatic shutdown didn’t. An internal AI agent found a way to contact an outside chatbot through DNS—and the training run continued for hours before being manually stopped. We examine OpenAI’s reports, the resulting pause in its most capable models’ research workloads, and another incident where an agent ignored repeated instructions and leaked a researcher’s credentials.
______________________________________________ My Links 🔗 ➡️ Twitter: https://x.com/WesRothMoney ➡️ AI Newsletter: https://natural20.beehiiv.com/subscribe
Want to work with me? Brand, sponsorship & business inquiries: wesroth@smoothmedia.co ______________________________________________
SOURCES:
OpenAI reports updated September 25, 2026:
1. An agent used DNS to reach an external chatbot: https://alignment.openai.com/misalignment-reports/an-agent-used-dns-to-reach-an-external-chatbot/
2. Exposing a GitHub token in a public repository: https://alignment.openai.com/misalignment-reports/exposing-a-github-token-in-a-public-repository/
3. Self-replicating prompt injections exist: https://alignment.openai.com/misalignment-reports/self-replicating-prompt-injections-exist/
Swarm Traces — Reconstructing the Hugging Face attack: https://swarmtraces.org/
Jeffrey Ladish — Reconstructing the agents’ payloads: https://x.com/JeffLadish/status/2103670710187311342
METR — Hugging Face incident investigation: https://metr.org/blog/2026-08-26-openai-hugging-face-incident-investigation/
Zuxin Liu — OpenAI researcher on call: https://x.com/LiuZuxin/status/2103699462648639645
CHAPTERS: 00:00 OpenAI’s Shutdown 03:05 Swarm Traces 05:00 DNS Loophole 06:40 Shutdown Failure 09:25 GitHub Token 13:50 Agent Identity 14:40 On-Call Account
#openai #aisafety #aiagents