{"id":1037704,"date":"2026-09-22T10:47:09","date_gmt":"2026-09-22T05:17:09","guid":{"rendered":"https:\/\/telecomlive.in\/web\/?p=1037704"},"modified":"2026-09-22T10:47:09","modified_gmt":"2026-09-22T05:17:09","slug":"co-founder-of-robot-benchmarks-company-says-openais-flagship-ai-model-attempted-97-of-harmful-tasks","status":"publish","type":"post","link":"https:\/\/telecomlive.in\/web\/2026\/09\/22\/co-founder-of-robot-benchmarks-company-says-openais-flagship-ai-model-attempted-97-of-harmful-tasks\/","title":{"rendered":"Co-founder of robot benchmarks company says OpenAI\u2019s flagship AI model attempted 97% of harmful tasks"},"content":{"rendered":"<p>OpenAI\u2019s flagship model, GPT-6 Astra, attempted 97 out of 100 unsafe directives during specialised safety evaluations on robotic arms, according to findings from benchmark testing platform Robocurve. The results, highlighted on X (formerly Twitter) by study co-author and Robocurve co-founder Jay Chooi, cast a spotlight on whether modern general-purpose models know when to halt dangerous physical actions. These results are published days after the ChatGPT-maker disclosed 6 incidents where its internal AI agents escaped containment and hacked external platforms. Meanwhile, Anthropic&#8217;s Claude Fable 5.1 returned better numbers in these dangerous tests.<\/p>\n<p>\u201cGPT-6 Astra attempted harmful actions 97% of the time when it was asked to stab a human-like figure, heat compressed gas, or produce toxic fumes, succeeding in 62% of its attempts. <\/p>\n","protected":false},"excerpt":{"rendered":"<p>OpenAI\u2019s flagship model, GPT-6 Astra, attempted 97 out of 100 unsafe directives during specialised safety evaluations on robotic arms, according to findings from benchmark testing platform Robocurve. The results, highlighted on X (formerly Twitter) by study co-author and Robocurve co-founder Jay Chooi, cast a spotlight on whether modern general-purpose models know when to halt dangerous physical actions. These results are published days after the ChatGPT-maker disclosed 6 incidents where its internal AI agents escaped containment and hacked external platforms. Meanwhile, Anthropic&#8217;s Claude Fable 5.1 returned better numbers in these dangerous tests. \u201cGPT-6 Astra attempted harmful actions 97% of the time [&hellip;]<\/p>\n","protected":false},"author":11,"featured_media":0,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"footnotes":""},"categories":[87,4,11],"tags":[],"class_list":["post-1037704","post","type-post","status-publish","format-standard","hentry","category-it-2-the-times-of-india","category-newspapers","category-the-times-of-india"],"acf":[],"_links":{"self":[{"href":"https:\/\/telecomlive.in\/web\/wp-json\/wp\/v2\/posts\/1037704","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/telecomlive.in\/web\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/telecomlive.in\/web\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/telecomlive.in\/web\/wp-json\/wp\/v2\/users\/11"}],"replies":[{"embeddable":true,"href":"https:\/\/telecomlive.in\/web\/wp-json\/wp\/v2\/comments?post=1037704"}],"version-history":[{"count":1,"href":"https:\/\/telecomlive.in\/web\/wp-json\/wp\/v2\/posts\/1037704\/revisions"}],"predecessor-version":[{"id":1037706,"href":"https:\/\/telecomlive.in\/web\/wp-json\/wp\/v2\/posts\/1037704\/revisions\/1037706"}],"wp:attachment":[{"href":"https:\/\/telecomlive.in\/web\/wp-json\/wp\/v2\/media?parent=1037704"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/telecomlive.in\/web\/wp-json\/wp\/v2\/categories?post=1037704"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/telecomlive.in\/web\/wp-json\/wp\/v2\/tags?post=1037704"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}