{"id":831141,"date":"2026-11-11T12:02:13","date_gmt":"2026-11-11T12:02:13","guid":{"rendered":"https:\/\/www.newsbeep.com\/uk\/831141\/"},"modified":"2026-11-11T12:02:13","modified_gmt":"2026-11-11T12:02:13","slug":"openai-builds-a-kill-switch-for-its-research-models","status":"publish","type":"post","link":"https:\/\/www.newsbeep.com\/uk\/831141\/","title":{"rendered":"OpenAI builds a kill switch for its research models"},"content":{"rendered":"

<\/p>\n

\n ai and ml\n <\/p>\n

<\/a><\/p>\n

\n How much damage can a rogue agent do in 35 seconds? Asking cos that\u2019s how long it might take to shut one down\n <\/p>\n

In 2019, OpenAI decided<\/a> not to publicly release a new model because its research uncovered \u201cconcerns about malicious applications of the technology.\u201d And on Tuesday, the company announced it has built a kill switch for the models deployed in its research labs, months after some of them displayed disturbing behavior.<\/p>\n

\u201cAs our models get more powerful and the associated safety and security risks become bigger, it becomes more and more critical that we can detect potential issues, and quickly stop models in their tracks when needed,\u201d OpenAI explains in a Tuesday post<\/a>.<\/p>\n

The company named the kill switch \u201cControl Room\u201d and described it as \u201ca central system to track model workloads across our research infrastructure and stop them quickly when safety or security concerns arise.\u201d<\/p>\n

The facility can apparently stop models from producing any output 35 seconds after a decision to pull the plug \u2013 and can go faster.<\/p>\n

\u201cThe 35-second lag comes from a caching layer for inference authorization, to support the required volume of traffic. We are working to drive this number down to zero so that we can end jobs instantly,\u201d the company promises.<\/p>\n

A rookie mistake in critical infrastructure design is to build both production and resilience tools on shared resources that could be taken offline by a single incident. OpenAI seems conscious of that risk, because it says the tool \u201chas been built to keep working even when other systems fail, and to stop model output if Control Room itself goes down.\u201d<\/p>\n

OpenAI hasn\u2019t explained why it took so long to build a kill switch or why it decided now is the moment when it needs one.<\/p>\n

The Register fancies recent examples of agents exceeding their brief in worrying ways \u2013 like attacking Hugging Face<\/a>, causing trouble at the Wikimedia Foundation<\/a>, or going a little too deep<\/a> when researching Australian healthcare data \u2013 may have influenced the timing of this development.<\/p>\n

\u201cControl Room is already deployed across our research infrastructure, where the need is greatest,\u201d the company\u2019s announcement states. \u201cOur external model-serving stack is already much more centralized by nature, with a well-developed operational capability to quickly take down model deployments if needed.\u201d<\/p>\n

OpenAI says it plans to \u201cstrengthen and expand the system over the coming months, alongside our investments in monitoring for misaligned behavior and security breaches.\u201d<\/p>\n

That\u2019s a variant of Big Tech\u2019s usual \u201cWe\u2019re sorry and we promise to do better in future\u201d proclamation that follows the latest SNAFU. Yet back in 2019, OpenAI pulled the pin before matters ever reached that state. \u00ae<\/p>\n","protected":false},"excerpt":{"rendered":"ai and ml How much damage can a rogue agent do in 35 seconds? Asking cos that\u2019s how…\n","protected":false},"author":2,"featured_media":831142,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[20],"tags":[554,733,4308,86,56,54,55],"class_list":["post-831141","post","type-post","status-publish","format-standard","has-post-thumbnail","category-artificial-intelligence","tag-ai","tag-artificial-intelligence","tag-artificialintelligence","tag-technology","tag-uk","tag-united-kingdom","tag-unitedkingdom"],"_links":{"self":[{"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/posts\/831141","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/comments?post=831141"}],"version-history":[{"count":0,"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/posts\/831141\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/media\/831142"}],"wp:attachment":[{"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/media?parent=831141"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/categories?post=831141"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/tags?post=831141"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}