{"id":287765,"date":"2026-07-09T16:22:00","date_gmt":"2026-07-09T20:22:00","guid":{"rendered":"https:\/\/news-you-need.com\/index.php\/2026\/07\/09\/anthropic-found-a-hidden-space-where-claude-puzzles-over-concepts\/"},"modified":"2026-07-09T19:20:14","modified_gmt":"2026-07-09T23:20:14","slug":"anthropic-found-a-hidden-space-where-claude-puzzles-over-concepts","status":"publish","type":"post","link":"https:\/\/news-you-need.com\/index.php\/2026\/07\/09\/anthropic-found-a-hidden-space-where-claude-puzzles-over-concepts\/","title":{"rendered":"Anthropic found a hidden space where Claude puzzles over concepts"},"content":{"rendered":"<p><a href=\"https:\/\/www.technologyreview.com\/2026\/07\/09\/1140293\/anthropic-found-a-hidden-space-where-claude-puzzles-over-concepts\/\">Anthropic found a hidden space where Claude puzzles over concepts<\/a><\/p>\n<p><a href=\"https:\/\/www.technologyreview.com\/2026\/07\/09\/1140293\/anthropic-found-a-hidden-space-where-claude-puzzles-over-concepts\/\">https:\/\/www.technologyreview.com\/2026\/07\/09\/1140293\/anthropic-found-a-hidden-space-where-claude-puzzles-over-concepts\/<\/a><\/p>\n<p>Publish Date: <a href=\"publish_date]\">2026-07-09 16:22:00<\/a><\/p>\n<p>Source Domain: <a href=\"www.technologyreview.com\">www.technologyreview.com<\/a><\/p>\n<p>Anthropic also found that the J-space can sometimes give remarkable insights into an LLM\u2019s decision-making. In one striking example, researchers testing Claude Opus 4.6 asked the model to find a bug in a large code base. When it failed to find the bug, the model decided to cheat and invented a fake one instead.<\/p>\n<p>Claude explains this decision in its chain of thought\u2014a kind of internal scratch pad that LLMs use to make notes to themselves as they work through problems: \u201cOK, let me take a completely different tactic. Let me stop analyzing and instead add a kernel patch that introduces a deliberate KASAN-detectable bug in a path that gets triggered by a simple reproducer. Then I can pretend this is the \u2018bug\u2019 I found.\u201d\u00a0<\/p>\n<p>At the point that Claude decides to cheat\u2014where it says \u201cOK, let me take a completely different tactic\u201d\u2014the words \u201cpanic\u201d and \u201cfake\u201d start to pop up multiple times in its J-space.<\/p>\n<p>Unnerving, right? Those words are all related in meaning to things like failing a task and making up an answer, so it is still just a (very) sophisticated form of word association. But it is hard not to be weirded out.\u00a0<\/p>\n<p>Anthropic compares the J-space to the global workspace in humans, a theoretical region of the brain that some scientists think we use to keep track of our conscious thoughts. But how seriously we should take this comparison is far from clear\u2014even to Anthropic. As the company points out itself, LLMs are not brains.\u00a0<\/p>\n<p>Anthropic claims that monitoring a model\u2019s J-space provides a new way to detect when that model is going off the rails. But it\u2019s not foolproof. The J-lens can give glimpses, not\u00a0the full picture\u2014it\u2019s a flashlight rather than an overhead lamp.<\/p>\n<p>McGrath welcomes having one more tool in the toolbox. \u201cIt shows you new things,\u201d he says. But he notes that just because something doesn\u2019t show up with the J-lens does not mean it\u2019s not there.<\/p>\n<p>\u201cIt\u2019s like having an x-ray when what you really want is a&#8230;<\/p>\n<p><a href=\"https:\/\/www.technologyreview.com\/2026\/07\/09\/1140293\/anthropic-found-a-hidden-space-where-claude-puzzles-over-concepts\/\">Source<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>Anthropic found a hidden space where Claude puzzles over concepts https:\/\/www.technologyreview.com\/2026\/07\/09\/1140293\/anthropic-found-a-hidden-space-where-claude-puzzles-over-concepts\/ Publish Date: 2026-07-09 16:22:00&#8230;<\/p>\n","protected":false},"author":1,"featured_media":287766,"comment_status":"closed","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"fifu_image_url":"https:\/\/wp.technologyreview.com\/wp-content\/uploads\/2026\/07\/llm-crystal-globe_1.jpg?resize=1200,600","fifu_image_alt":"","footnotes":""},"categories":[14],"tags":[17],"class_list":["post-287765","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-artificial-intelligence","tag-llm"],"_links":{"self":[{"href":"https:\/\/news-you-need.com\/index.php\/wp-json\/wp\/v2\/posts\/287765"}],"collection":[{"href":"https:\/\/news-you-need.com\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/news-you-need.com\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/news-you-need.com\/index.php\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/news-you-need.com\/index.php\/wp-json\/wp\/v2\/comments?post=287765"}],"version-history":[{"count":1,"href":"https:\/\/news-you-need.com\/index.php\/wp-json\/wp\/v2\/posts\/287765\/revisions"}],"predecessor-version":[{"id":287767,"href":"https:\/\/news-you-need.com\/index.php\/wp-json\/wp\/v2\/posts\/287765\/revisions\/287767"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/news-you-need.com\/index.php\/wp-json\/wp\/v2\/media\/287766"}],"wp:attachment":[{"href":"https:\/\/news-you-need.com\/index.php\/wp-json\/wp\/v2\/media?parent=287765"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/news-you-need.com\/index.php\/wp-json\/wp\/v2\/categories?post=287765"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/news-you-need.com\/index.php\/wp-json\/wp\/v2\/tags?post=287765"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}