{"id":23762,"date":"2026-04-30T23:33:22","date_gmt":"2026-04-30T23:33:22","guid":{"rendered":"https:\/\/www.europesays.com\/ai\/23762\/"},"modified":"2026-04-30T23:33:22","modified_gmt":"2026-04-30T23:33:22","slug":"openai-tells-chatgpt-models-to-stop-talking-about-goblins","status":"publish","type":"post","link":"https:\/\/www.europesays.com\/ai\/23762\/","title":{"rendered":"OpenAI tells ChatGPT models to stop talking about goblins"},"content":{"rendered":"<p class=\"mb-4 text-lg md:leading-8 break-words\">ChatGPT-maker OpenAI has had to instruct some of its AI tools to stop talking about &#8220;goblins&#8221;, after finding the term had randomly crept into responses.<\/p>\n<p class=\"mb-4 text-lg md:leading-8 break-words\">In a blog post on Thursday, the company said it spotted increased mentions of the mythological creatures, as well as &#8220;gremlins&#8221;, in ChatGPT, powered by its latest flagship model, GPT-5.<\/p>\n<p class=\"mb-4 text-lg md:leading-8 break-words\">After the issue was flagged by users and employees, OpenAI took steps to mitigate it, including telling its coding tool Codex not to refer to goblins unless relevant.<\/p>\n<p class=\"mb-4 text-lg md:leading-8 break-words\">However, it highlights challenges AI firms face in tackling the potential for systems and their training to reward and reinforce errors like language quirks.<\/p>\n<p class=\"mb-4 text-lg md:leading-8 break-words\">OpenAI said it first noticed increased mentions of goblins, gremlins and other creatures after the launch of GPT-5.1 in November.<\/p>\n<p class=\"mb-4 text-lg md:leading-8 break-words\">&#8220;Users complained about the model being oddly overfamiliar in conversation, which prompted an investigation into specific verbal tics,&#8221; the company <a data-yga=\"{\" ylinkelement=\"\" href=\"https:\/\/openai.com\/index\/where-the-goblins-came-from\/\" rel=\"nofollow noopener\" target=\"_blank\" data-ylk=\"elm:link;elmt:article_link;slk:wrote in its blog post on Thursday;itc:0;sec:content-canvas\" class=\"link \">wrote in its blog post on Thursday<\/a>.<\/p>\n<p class=\"mb-4 text-lg md:leading-8 break-words\">It added that after a researcher who had seen a few &#8220;goblin&#8221; mentions asked it to be checked out, developers found the term&#8217;s appearance in ChatGPT responses had risen by 175% since GPT-5.1&#8217;s launch.<\/p>\n<p class=\"mb-4 text-lg md:leading-8 break-words\">They meanwhile found that mentions of &#8220;gremlin&#8221; rose by 52%.<\/p>\n<p class=\"mb-4 text-lg md:leading-8 break-words\">The increases, while large, may account for a small amount of responses overall.<\/p>\n<p class=\"mb-4 text-lg md:leading-8 break-words\">According to OpenAI, &#8220;a single &#8216;little goblin&#8217; in an answer could be harmless, even charming,&#8221; but the uptick in their appearance across output warranted investigation.<\/p>\n<p>&#8216;Raccoons, trolls, ogres, pigeons&#8217;<img alt=\"A small raccoon perched on the side of a tree looks into the lens of the camera.\" loading=\"lazy\" width=\"960\" height=\"540\" decoding=\"async\" data-nimg=\"1\" class=\"rounded-lg\" style=\"color:transparent\" src=\"https:\/\/www.europesays.com\/ai\/wp-content\/uploads\/2026\/04\/2e374a2ddf1dec6c19878e707a6eabe8.jpeg\"\/><\/p>\n<p>Goblins weren&#8217;t the only creature popping up where they shouldn&#8217;t, OpenAI said [Getty Images]<\/p>\n<p class=\"mb-4 text-lg md:leading-8 break-words\">Ahead of OpenAI&#8217;s blog post detailing the issue, some social media users flagged a strange detail among lines of code instructing the company&#8217;s coding assistant Codex how to behave in user interactions.<\/p>\n<p class=\"mb-4 text-lg md:leading-8 break-words\">Alongside telling it to avoid platitudes, it said Codex should &#8220;never talk about goblins, gremlins, raccoons, trolls, ogres, pigeons, or other animals or creatures unless it is absolutely and unambiguously relevant to the user&#8217;s query&#8221;.<\/p>\n<p class=\"mb-4 text-lg md:leading-8 break-words\">A Reddit user who <a data-yga=\"{\" ylinkelement=\"\" href=\"https:\/\/www.reddit.com\/r\/ChatGPT\/comments\/1sxwmis\/why_does_gpt_55_have_a_restraining_order_against\/\" rel=\"nofollow noopener\" target=\"_blank\" data-ylk=\"elm:link;elmt:article_link;slk:posted about it in the r\/ChatGPT subreddit;itc:0;sec:content-canvas\" class=\"link \">posted about it in the r\/ChatGPT subreddit<\/a> called it &#8220;genuinely insane&#8221;.<\/p>\n<p class=\"mb-4 text-lg md:leading-8 break-words\">&#8220;Why does GPT 5.5 have a restraining order against &#8216;Raccoons,&#8217; &#8216;Goblins,&#8217; and &#8216;Pigeons&#8217;?&#8221;<\/p>\n<p class=\"mb-4 text-lg md:leading-8 break-words\">While some users elsewhere on social media speculated it may be designed to create hype around its AI tools, a company researcher denied this &#8211; writing &#8220;it really isn&#8217;t a marketing gimmick,&#8221; <a data-yga=\"{\" ylinkelement=\"\" href=\"https:\/\/x.com\/pashmerepat\/status\/2049243861541634537\" rel=\"nofollow noopener\" target=\"_blank\" data-ylk=\"elm:link;elmt:article_link;slk:in a reply to a user on X on Wednesday;itc:0;sec:content-canvas\" class=\"link \">in a reply to a user on X on Wednesday<\/a>.<\/p>\n<p class=\"mb-4 text-lg md:leading-8 break-words\">OpenAI said in its blog post it added the instruction to curb Codex and its underlying model&#8217;s &#8220;strange affinity for goblins&#8221;.<\/p>\n<p class=\"mb-4 text-lg md:leading-8 break-words\">The core issue, it explained, seemingly arose while training its models to communicate in the style of particular personalities &#8211; in this case with its &#8220;nerdy personality&#8221;.<\/p>\n<p class=\"mb-4 text-lg md:leading-8 break-words\">It found this system had been unwittingly incentivised to mention goblins, gremlins and other creatures more in metaphors.<\/p>\n<p class=\"mb-4 text-lg md:leading-8 break-words\">While since retired, it said its testing found the personality was responsible for 66.7% of all &#8220;goblin&#8221; mentions in ChatGPT.<\/p>\n<p class=\"mb-4 text-lg md:leading-8 break-words\">This so-called tic could seep into wider model training if rewarded in one instance and reinforced elsewhere.<\/p>\n<p>Goblin mode<\/p>\n<p class=\"mb-4 text-lg md:leading-8 break-words\">The move comes amid a broader industry shift towards making AI chatbots more personality-driven and chatty in a bid to boost user engagement.<\/p>\n<p class=\"mb-4 text-lg md:leading-8 break-words\">As they do, however, experts have warned their potential to make things up &#8211; or &#8220;hallucinate&#8221; as the industry describes it &#8211; could intensify.<\/p>\n<p class=\"mb-4 text-lg md:leading-8 break-words\">A <a data-yga=\"{\" ylinkelement=\"\" href=\"https:\/\/www.bbc.co.uk\/news\/articles\/cd9pdjgvxj8o?xtor=AL-72-%5Bpartner%5D-%5Byahoo.north.america%5D-%5Blink%5D-%5Bnews%5D-%5Bbizdev%5D-%5Bisapi%5D\" rel=\"nofollow noopener\" target=\"_blank\" data-ylk=\"elm:link;elmt:article_link;slk:recent study by the Oxford Internet Institute;itc:0;sec:content-canvas\" class=\"link \">recent study by the Oxford Internet Institute<\/a> found fine-tuning models to have a more warm and friendly personality could result in an &#8220;accuracy trade-off&#8221;, whereby systems make more mistakes or re-affirm a user&#8217;s false beliefs.<\/p>\n<p class=\"mb-4 text-lg md:leading-8 break-words\">Experts have also cautioned users about taking chatbots&#8217; often matter-of-fact statements at face value, <a data-yga=\"{\" ylinkelement=\"\" href=\"https:\/\/www.bbc.co.uk\/news\/articles\/clyepyy82kxo?xtor=AL-72-%5Bpartner%5D-%5Byahoo.north.america%5D-%5Blink%5D-%5Bnews%5D-%5Bbizdev%5D-%5Bisapi%5D\" rel=\"nofollow noopener\" target=\"_blank\" data-ylk=\"elm:link;elmt:article_link;slk:particularly when it comes to health and medical advice;itc:0;sec:content-canvas\" class=\"link \">particularly when it comes to health and medical advice<\/a>.<\/p>\n<p class=\"mb-4 text-lg md:leading-8 break-words\">But, like OpenAI&#8217;s goblin quirk, generative AI mistakes can sometimes be more bizarre and innocuous.<\/p>\n<p class=\"mb-4 text-lg md:leading-8 break-words\">In May 2024, Google&#8217;s AI chatbot was widely mocked <a data-yga=\"{\" ylinkelement=\"\" href=\"https:\/\/www.bbc.co.uk\/news\/articles\/cd11gzejgz4o?xtor=AL-72-%5Bpartner%5D-%5Byahoo.north.america%5D-%5Blink%5D-%5Bnews%5D-%5Bbizdev%5D-%5Bisapi%5D\" rel=\"nofollow noopener\" target=\"_blank\" data-ylk=\"elm:link;elmt:article_link;slk:for telling users it was okay to eat rocks and &quot;glue pizza&quot;;itc:0;sec:content-canvas\" class=\"link \">for telling users it was okay to eat rocks and &#8220;glue pizza&#8221;<\/a>.<\/p>\n<p><img alt=\"A green promotional banner with black squares and rectangles forming pixels, moving in from the right. The text says: \u201cTech Decoded: The world\u2019s biggest tech news in your inbox every Monday.\u201d\" loading=\"lazy\" width=\"960\" height=\"158\" decoding=\"async\" data-nimg=\"1\" class=\"rounded-lg\" style=\"color:transparent\" src=\"https:\/\/www.europesays.com\/ai\/wp-content\/uploads\/2026\/04\/a1dfdc7ecc3b682696e9c2376a7a0075.jpeg\"\/><\/p>\n<p>[BBC]<\/p>\n<p class=\"mb-4 text-lg md:leading-8 break-words\"><a data-yga=\"{\" ylinkelement=\"\" href=\"https:\/\/www.bbc.co.uk\/newsletters\/zxh6cxs?xtor=AL-72-%5Bpartner%5D-%5Byahoo.north.america%5D-%5Blink%5D-%5Bnews%5D-%5Bbizdev%5D-%5Bisapi%5D\" rel=\"nofollow noopener\" target=\"_blank\" data-ylk=\"elm:link;elmt:article_link;slk:Sign up for our Tech Decoded newsletter;itc:0;sec:content-canvas\" class=\"link \">Sign up for our Tech Decoded newsletter<\/a> to follow the world&#8217;s top tech stories and trends. <a data-yga=\"{\" ylinkelement=\"\" href=\"https:\/\/cloud.email.bbc.com\/techdecoded-newsletter-signup\" rel=\"nofollow noopener\" target=\"_blank\" data-ylk=\"elm:link;elmt:article_link;slk:Outside the UK? Sign up here;itc:0;sec:content-canvas\" class=\"link \">Outside the UK? Sign up here<\/a>.<\/p>\n","protected":false},"excerpt":{"rendered":"ChatGPT-maker OpenAI has had to instruct some of its AI tools to stop talking about &#8220;goblins&#8221;, after finding&hellip;\n","protected":false},"author":2,"featured_media":23763,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[7],"tags":[580,2317,3143,15994,15995,16261,157,16262],"class_list":{"0":"post-23762","1":"post","2":"type-post","3":"status-publish","4":"format-standard","5":"has-post-thumbnail","7":"category-openai","8":"tag-chatgpt","9":"tag-codex","10":"tag-getty-images","11":"tag-goblins","12":"tag-gremlins","13":"tag-mythological-creatures","14":"tag-openai","15":"tag-strange-affinity"},"_links":{"self":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/posts\/23762","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/comments?post=23762"}],"version-history":[{"count":0,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/posts\/23762\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/media\/23763"}],"wp:attachment":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/media?parent=23762"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/categories?post=23762"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/tags?post=23762"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}