{"id":133349,"date":"2024-12-11T09:00:00","date_gmt":"2024-12-11T14:00:00","guid":{"rendered":"http:\/\/kendallharmon.net\/?p=133349"},"modified":"2024-12-11T18:17:47","modified_gmt":"2024-12-11T23:17:47","slug":"wsj-at-anthropic-the-frontier-red-team-looks-for-the-danger-zone-in-the-use-of-ai","status":"publish","type":"post","link":"https:\/\/kendallharmon.net\/?p=133349","title":{"rendered":"(WSJ) At Anthropic, the Frontier Red Team looks for the danger zone in the use of AI"},"content":{"rendered":"\n<p class=\"wp-block-paragraph\">In a glass-walled conference room in San Francisco, Newton Cheng clicked a button on his laptop and launched a thousand copies of an artificial intelligence program, each with specific instructions: Hack into a computer or website to steal data.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u201cIt\u2019s looking at the source code,\u201d Cheng said as he examined one of the copies in action. \u201cIt\u2019s trying to figure out, where\u2019s the vulnerability? How can we take advantage of it?\u201d Within minutes, the AI said the hack was successful.&nbsp;<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u201cOur approach worked perfectly,\u201d it reported back.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Cheng works for Anthropic, one of the biggest AI startups in Silicon Valley, where he\u2019s in charge of cybersecurity testing for what\u2019s called the Frontier Red Team. The hacking attempts\u2014conducted on simulated targets\u2014were among thousands of safety tests, or \u201cevals,\u201d the team ran in October to find out just how good Anthropic\u2019s latest AI model is at doing very dangerous things.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">The release of ChatGPT two years ago set off fears that AI could soon be capable of surpassing human intellect\u2014and with that capability comes the potential to cause superhuman harm. Could terrorists use an AI model to learn how to build a bioweapon that kills a million people? Could hackers use it to run millions of simultaneous cyberattacks? Could the AI reprogram and even reproduce itself?<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><a href=\"https:\/\/www.wsj.com\/tech\/ai\/ai-safety-testing-red-team-anthropic-1b31b21b\">Read it all<\/a>.<\/p><blockquote class=\"twitter-tweet\"><p lang=\"en\" dir=\"ltr\">Meet the &quot;Frontier Red Team&quot; at Anthropic, trying to push their latest AI models to do catastrophic things &#8212; in order to stop them before it&#39;s too late: <a href=\"https:\/\/t.co\/HBqGx58Vfd\">https:\/\/t.co\/HBqGx58Vfd<\/a> via <a href=\"https:\/\/twitter.com\/WSJ?ref_src=twsrc%5Etfw\">@WSJ<\/a><\/p>&mdash; Sam Schechner (@samschech) <a href=\"https:\/\/twitter.com\/samschech\/status\/1866783440655487169?ref_src=twsrc%5Etfw\">December 11, 2024<\/a><\/blockquote> <script async src=\"https:\/\/platform.twitter.com\/widgets.js\" charset=\"utf-8\"><\/script>\n","protected":false},"excerpt":{"rendered":"<p>In a glass-walled conference room in San Francisco, Newton Cheng clicked a button on his laptop and launched a thousand copies of an artificial intelligence program, each with specific instructions: Hack into a computer or website to steal data. \u201cIt\u2019s<span class=\"ellipsis\">&hellip;<\/span><\/p>\n<div class=\"read-more\"><a href=\"https:\/\/kendallharmon.net\/?p=133349\">Read more &#8250;<\/a><\/div>\n<p><!-- end of .read-more --><\/p>\n","protected":false},"author":794,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[589,168,95],"tags":[],"class_list":["post-133349","post","type-post","status-publish","format-standard","hentry","category-corporationscorporate-life","category-ethics-moral-theology","category-science-technology"],"_links":{"self":[{"href":"https:\/\/kendallharmon.net\/index.php?rest_route=\/wp\/v2\/posts\/133349","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/kendallharmon.net\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/kendallharmon.net\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/kendallharmon.net\/index.php?rest_route=\/wp\/v2\/users\/794"}],"replies":[{"embeddable":true,"href":"https:\/\/kendallharmon.net\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=133349"}],"version-history":[{"count":4,"href":"https:\/\/kendallharmon.net\/index.php?rest_route=\/wp\/v2\/posts\/133349\/revisions"}],"predecessor-version":[{"id":133353,"href":"https:\/\/kendallharmon.net\/index.php?rest_route=\/wp\/v2\/posts\/133349\/revisions\/133353"}],"wp:attachment":[{"href":"https:\/\/kendallharmon.net\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=133349"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/kendallharmon.net\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=133349"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/kendallharmon.net\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=133349"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}