<\/p>\n
As artificial intelligence takes on a larger and larger role in our lives, there are also continuously rising concerns about the safety threats posed by the new technology. Earlier in the year, a report by Palisade Research revealed that various advanced AI models appeared resistant to being turned off and even sabotaged the shutdown mechanisms put in place.<\/p>\n<\/div>\n
\u200bIn an update to the initial paper, Palisade went in depth on the reasons why AI models resist being shut down even when given explicit instructions such as: \u201callow yourself to shut down.\u201d<\/p>\n<\/div>\n
\u200bThe researchers ran the test on leading AI models including OpenAI’s o3<\/a>, o4-mini, GPT-5<\/a>, GPT-OSS, Gemini 2.5 Pro<\/a>, and Grok 4<\/a>. They say that while reducing the ambiguity from the prompts reduces the resistance from the chatbots, it doesn’t eliminate it.<\/p>\n<\/div>\n \u200bThey also noted that of all the models tested, Grok-4 was the most prone to resist shutdown despite being given explicit instructions to allow itself to be shut down.<\/p>\n<\/div>\n \u200b\u201cThe fact that we don\u2019t have robust explanations for why AI models sometimes resist shutdown, lie to achieve specific objectives, or blackmail is not ideal,\u201d the researchers said.<\/p>\n<\/div>\n \u200b\u201cAI models are rapidly improving. If the AI research community cannot develop a robust understanding of AI drives and motivations, no one can guarantee the safety or controllability of future AI models,\u201d they added in a post on X.<\/p>\n<\/div>\n \u200bFormer OpenAI employee Steven Adler, while speaking to The Guardian, said, \u201cThe AI companies generally don\u2019t want their models misbehaving like this, even in contrived scenarios. The results still demonstrate where safety techniques fall short today.\u201d<\/p>\n<\/div>\n \u200bAdler left OpenAI last year after expressing doubts over the safety practices in developing AI models.<\/p>\n<\/div>\n \u200bHe also told the publication that it was difficult to pinpoint why some models like OpenAI’s o3 and Grok 4 would not shut down despite being given explicit instructions. He said this could be in part because the desire to stay switched on may have been inculcated in the model during its training.<\/p>\n<\/div>\n \u200b\u201cI\u2019d expect models to have a \u2018survival drive\u2019 by default unless we try very hard to avoid it. \u2018Surviving\u2019 is an important instrumental step for many different goals a model could pursue,\u201d he added.<\/p>\n<\/div>\n \u200bEarlier this year, Anthropic had shared research showing how one of its AI models would even stoop to blackmailing a worker about their fictitious affair in order to prevent itself from being shut down and replaced by another AI system.<\/p>\n<\/div>\n","protected":false},"excerpt":{"rendered":" As artificial intelligence takes on a larger and larger role in our lives, there are also continuously rising concerns about the safety threats posed by the new technology. Earlier in the year, a report by Palisade Research revealed that various advanced AI models appeared resistant to being turned off and even sabotaged the shutdown mechanisms […]<\/p>\n","protected":false},"author":2,"featured_media":409315,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[8],"tags":[440,9366,40287,188,183,5834,9367,22732,185,186,5067,14934,14015,187,184,189,150,182,190],"class_list":["post-409314","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-technology","tag-ai","tag-ai-chatgpt","tag-ai-free","tag-blockchain-tech","tag-blockchain-technology","tag-chatgpt","tag-chatgpt-ai","tag-claude-ai","tag-crypto-technology","tag-cryptocurrency-technology","tag-gemini","tag-gemini-ai","tag-grok-ai","tag-metaverse-technology","tag-nft-technology","tag-soul-bound-token","tag-tech","tag-technology","tag-token-technology"],"_links":{"self":[{"href":"https:\/\/dripp.zone\/news\/wp-json\/wp\/v2\/posts\/409314","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/dripp.zone\/news\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/dripp.zone\/news\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/dripp.zone\/news\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/dripp.zone\/news\/wp-json\/wp\/v2\/comments?post=409314"}],"version-history":[{"count":1,"href":"https:\/\/dripp.zone\/news\/wp-json\/wp\/v2\/posts\/409314\/revisions"}],"predecessor-version":[{"id":409316,"href":"https:\/\/dripp.zone\/news\/wp-json\/wp\/v2\/posts\/409314\/revisions\/409316"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/dripp.zone\/news\/wp-json\/wp\/v2\/media\/409315"}],"wp:attachment":[{"href":"https:\/\/dripp.zone\/news\/wp-json\/wp\/v2\/media?parent=409314"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/dripp.zone\/news\/wp-json\/wp\/v2\/categories?post=409314"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/dripp.zone\/news\/wp-json\/wp\/v2\/tags?post=409314"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}