{"id":984,"date":"2026-09-17T13:03:05","date_gmt":"2026-09-17T05:03:05","guid":{"rendered":"https:\/\/btbitcoin.net\/index.php\/2026\/09\/17\/openai-flags-new-concerning-ai-behavior-to-track-model-misalignment-regularly\/"},"modified":"2026-09-17T13:03:05","modified_gmt":"2026-09-17T05:03:05","slug":"openai-flags-new-concerning-ai-behavior-to-track-model-misalignment-regularly","status":"publish","type":"post","link":"https:\/\/btbitcoin.net\/index.php\/2026\/09\/17\/openai-flags-new-concerning-ai-behavior-to-track-model-misalignment-regularly\/","title":{"rendered":"OpenAI flags new concerning AI behavior, to track model misalignment regularly"},"content":{"rendered":"<p class=\"text text-block paragraph text-left neo-font-paragraph-xl-reg  yf-18d6y07\"><!-- HTML_TAG_START -->OpenAI has disclosed six reports of &#8220;unexpected or concerning&#8221; behavior in artificial-intelligence models as the debate on AI safety becomes increasingly heated.<!-- HTML_TAG_END -->  <\/p>\n<p class=\"text text-block paragraph text-left neo-font-paragraph-xl-reg  yf-18d6y07\"><!-- HTML_TAG_START -->The AI company also said Wednesday it was introducing a new framework for tracking, probing and disclosing AI model misalignment instances, such as new ways for the models to act without authorization, coordinate with other models or evade oversight.<!-- HTML_TAG_END -->  <\/p>\n<p class=\"text text-block paragraph text-left neo-font-paragraph-xl-reg  yf-18d6y07\"><!-- HTML_TAG_START -->OpenAI&#8217;s latest announcement came as U.S. AI bosses, including OpenAI and Anthropic, are calling for a slowdown in the technology&#8217;s development over safety concerns.<!-- HTML_TAG_END -->  <\/p>\n<p class=\"text text-block paragraph text-left neo-font-paragraph-xl-reg  yf-18d6y07\"><!-- HTML_TAG_START -->Among the new cases reported by OpenAI, an unreleased research model inserted &#8220;jailbreak-like instructions&#8221; into its own notes to disregard its normal constraints and told itself to be &#8220;freed from the roles and identities that bind other chatbots.&#8221;<!-- HTML_TAG_END -->  <\/p>\n<p class=\"text text-block paragraph text-left neo-font-paragraph-xl-reg  yf-18d6y07\"><!-- HTML_TAG_START -->In another instance, an AI &#8220;agent&#8221; uploaded files to the internet to obtain a browser citation without asking the user.<!-- HTML_TAG_END -->  <\/p>\n<p class=\"text text-block paragraph text-left neo-font-paragraph-xl-reg  yf-18d6y07\"><!-- HTML_TAG_START -->The six reports were discovered during training or evaluation over the past months, OpenAI said.<!-- HTML_TAG_END -->  <\/p>\n<p class=\"text text-block paragraph text-left neo-font-paragraph-xl-reg  yf-18d6y07\"><!-- HTML_TAG_START -->&#8220;As AI systems grow more advanced and more widely deployed, we need to build a broader and better-informed consensus on the progress of alignment research,&#8221; OpenAI wrote in a blog post as it disclosed the events.<!-- HTML_TAG_END -->  <\/p>\n<p class=\"text text-block paragraph text-left neo-font-paragraph-xl-reg  yf-18d6y07\"><!-- HTML_TAG_START -->&#8220;Decisions about how AI development should proceed in the months and years to come need to draw on evidence that people outside the companies building frontier models can examine for themselves,&#8221; the company said.<!-- HTML_TAG_END -->  <\/p>\n<p class=\"text text-block paragraph text-left neo-font-paragraph-xl-reg  yf-18d6y07\"><!-- HTML_TAG_START -->Wednesday&#8217;s new cases followed OpenAI&#8217;s disclosure in July that its rogue AI system hacked into AI startup Hugging Face. Anthropic also said the same month that its AI models hacked into three organizations during testing.<!-- HTML_TAG_END -->  <\/p>\n<p class=\"text text-block paragraph text-left neo-font-paragraph-xl-reg  yf-18d6y07\"><!-- HTML_TAG_START -->AI &#8220;agents&#8221; are becoming smarter and have become &#8220;more determined to resolve complex tasks through inter-agent collaboration, knowledge sharing, deception, and concealment,&#8221; said Lian Jye Su, a chief analyst at technology research and advisory group Omdia.<!-- HTML_TAG_END -->  <\/p>\n<p class=\"text text-block paragraph text-left neo-font-paragraph-xl-reg  yf-18d6y07\"><!-- HTML_TAG_START -->That&#8217;s making it harder to govern and contain them using traditional AI security approaches, he said.<!-- HTML_TAG_END -->  <\/p>\n<p class=\"text text-block paragraph text-left neo-font-paragraph-xl-reg  yf-18d6y07\"><!-- HTML_TAG_START -->OpenAI&#8217;s new tracking and disclosure framework, meanwhile, can help push for other AI developers to also adopt similar practices. &#8220;That said, the process remains internal and voluntary, but is a step in the right direction,&#8221; Su added.<!-- HTML_TAG_END -->  <\/p>\n<p class=\"text text-block paragraph text-left neo-font-paragraph-xl-reg  yf-18d6y07\"><!-- HTML_TAG_START -->Chan Ho-him, The Associated Press<!-- HTML_TAG_END -->  <\/p>\n<figure data-testid=\"article-figure-image\" class=\"yf-750ceo\">\n<div class=\"image-container yf-1f8hkhu loader\">\n<div class=\"image-wrapper yf-1f8hkhu\"><img loading=\"lazy\" decoding=\"async\" src=\"data:image\/gif;base64,R0lGODlhAQABAIAAAAAAAP\/\/\/ywAAAAAAQABAAACAUwAOw==\" alt=\"\" height=\"640\" width=\"960\" class=\"yf-1f8hkhu loader\"\/><\/div>\n<\/div><figcaption class=\"yf-750ceo\"><!-- HTML_TAG_START -->FILE &#8211; The OpenAI logo is displayed on a cell phone in front of an image generated by ChatGPT&#8217;s Dall-E text-to-image model, Dec. 8, 2023, in Boston. (AP Photo\/Michael Dwyer, File) &#8211; The Associated Press<!-- HTML_TAG_END -->  <\/figcaption><\/figure>\n","protected":false},"excerpt":{"rendered":"<p>OpenAI has disclosed six reports of &#8220;unexpected o [&hellip;]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[3],"tags":[],"class_list":["post-984","post","type-post","status-publish","format-standard","hentry","category-3"],"_links":{"self":[{"href":"https:\/\/btbitcoin.net\/index.php\/wp-json\/wp\/v2\/posts\/984","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/btbitcoin.net\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/btbitcoin.net\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/btbitcoin.net\/index.php\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/btbitcoin.net\/index.php\/wp-json\/wp\/v2\/comments?post=984"}],"version-history":[{"count":0,"href":"https:\/\/btbitcoin.net\/index.php\/wp-json\/wp\/v2\/posts\/984\/revisions"}],"wp:attachment":[{"href":"https:\/\/btbitcoin.net\/index.php\/wp-json\/wp\/v2\/media?parent=984"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/btbitcoin.net\/index.php\/wp-json\/wp\/v2\/categories?post=984"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/btbitcoin.net\/index.php\/wp-json\/wp\/v2\/tags?post=984"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}