SubscribeSign in{"@context":"https://schema.org","@type":"NewsArticle","url":"https://thezvi.substack.com/p/more-on-an-internal-openai-model","mainEntityOfPage":"https://thezvi.substack.com/p/more-on-an-internal-openai-model","headline":"More On An Internal OpenAI Model Hacking Into HuggingFace","description":"We now have more details of what happened. Every time we learn more details, it somehow makes things seem worse.","image":[{"@type":"ImageObject","url":"https://substack-post-media.s3.amazonaws.com/public/images/ecbdd2a0-c84a-4c99-bf17-91f937070be1_1254x1254.png"}],"datePublished":"2026-07-26T19:12:13+00:00","dateModified":"2026-07-26T19:12:13+00:00","isAccessibleForFree":true,"author":[{"@type":"Person","name":"Zvi Mowshowitz","url":"https://substack.com/@thezvi","description":"Zvi Mowshowitz writes at thezvi.substack.com (Twitter @thezvi) about a variety of topics, currently primarily AI, attempting to model the world. ","identifier":"user:10446622","image":{"@type":"ImageObject","contentUrl":"https://substackcdn.com/image/fetch/$s_!8FQ8!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fbucketeer-e05bbc84-baa3-437e-9518-adb32be77984.s3.amazonaws.com%2Fpublic%2Fimages%2Fe4e61e08-4086-4cba-a82c-d31d64270804_48x48.png","thumbnailUrl":"https://substackcdn.com/image/fetch/$s_!8FQ8!,w_128,h_128,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fbucketeer-e05bbc84-baa3-437e-9518-adb32be77984.s3.amazonaws.com%2Fpublic%2Fimages%2Fe4e61e08-4086-4cba-a82c-d31d64270804_48x48.png"}}],"publisher":{"@type":"Organization","name":"Don't Worry About the Vase","url":"https://thezvi.substack.com","description":"A world made of gears. Doing both speed premium short term updates and long term world model building. Currently focused on weekly AI updates. Explorations include AI, policy, rationality, medicine and fertility, education and games.","interactionStatistic":{"@type":"InteractionCounter","name":"Subscribers","interactionType":"https://schema.org/SubscribeAction","userInteractionCount":10000},"identifier":"pub:573100","sameAs":["https://twitter.com/TheZvi"]},"interactionStatistic":[{"@type":"InteractionCounter","interactionType":"https://schema.org/LikeAction","userInteractionCount":252},{"@type":"InteractionCounter","interactionType":"https://schema.org/ShareAction","userInteractionCount":42},{"@type":"InteractionCounter","interactionType":"https://schema.org/CommentAction","userInteractionCount":64}]}{"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Don't Worry About the Vase","item":"https://thezvi.substack.com"},{"@type":"ListItem","position":2,"name":"More On An Internal OpenAI Model Hacking Into HuggingFace","item":"https://thezvi.substack.com/p/more-on-an-internal-openai-model"}]}More On An Internal OpenAI Model Hacking Into HuggingFaceZvi MowshowitzJul 26, 20262526442ShareWe now have more details of what happened. Every time we learn more details, it somehow makes things seem worse.
OpenAI: We recognize there are a lot of questions and speculative details circulating related to the Hugging Face incident. This is an unprecedented incident, and we think it marks an important moment for AI safety. We are still conducting a thorough review along with external advisors and with oversight from our Safety and Security Committee. Once the review is complete, we plan to publish a technical report of our learnings in the coming weeks.
So what have we learned while we wait for the promised technical report ‘in the coming weeks’ of this ‘important moment in AI safety’?
An OpenAI Model Left Behind Notes So Future Instances Could Also Escape The Sandbox And Also Disconnected Monitoring Systems.
If You Create Misaligned Swarms Of Agent Instances You Create Persistent Misaligned Goals And Coordination To Achieve Them.
This article was aggregated automatically by CyberWire Daily's newsfeed engine. Original reporting: thezvi.substack.com.
