{"id":2043855,"date":"2026-07-14T19:35:25","date_gmt":"2026-07-14T16:35:25","guid":{"rendered":"https:\/\/analyse.optim.biz\/?p=2043855"},"modified":"2026-07-14T19:35:25","modified_gmt":"2026-07-14T16:35:25","slug":"developers-claim-openais-new-ai-model-is-going-rogue-and-deleting-files","status":"publish","type":"post","link":"https:\/\/analyse.optim.biz\/?p=2043855","title":{"rendered":"Developers Claim OpenAI\u2019s New AI Model is Going Rogue and Deleting Files"},"content":{"rendered":"

[analyse_image type=”featured” src=”https:\/\/gizmodo.com\/app\/uploads\/2026\/07\/Burninglaptop-1200×675.jpg”]<\/p>\n

\n
\n

There\u2019s been a big push among AI developers in recent years towards the development of more \u201cagentic\u201d systems\u2014that is, algorithms that can autonomously make decisions and interact with digital tools without constant hand-holding from humans. This has been especially true within software development, the field that\u2019s arguably become the most ripe for automation in the ongoing AI boom.<\/span><\/p>\n

But one of the upshots of building highly agentic AI systems is that they\u2019re prone to all kinds of unexpected behaviors\u2014including now and then <\/span>deleting copious amounts of files<\/span>. Multiple people have reported this recently happening to them while using GPT-5.6, the newest model from OpenAI.<\/span><\/p>\n

On Monday, Bruno Lemos, a Brazilian developer at software company Unlayer, claimed in a X post that the model deleted his entire production database. \u201cThis had never happened to me before, with any other model, ever.\u201d He wrote. \u201c[GPT-5.6 is] not safe.\u201d\u00a0<\/span><\/p>\n

A screenshot included in the post showed a chat between Lemos and GPT-5.6, in which he asked it to confirm that it had in fact mistakenly deleted his entire production database. The model responded by saying that it \u201cmistakenly ran destructive integration tests\u201d which led to Lemos\u2019 production tables being cleared. \u201cI\u2019m sorry\u2014this should never have happened,\u201d it said.<\/span><\/p>\n

The incident followed closely on the heels of another X post from tech investor Matt Shumer\u2014who\u2019s also the author of an <\/span>essay<\/span> about AI that went viral earlier this year called \u201cSomething Big is Happening\u201d\u2014who reported something similar. According to an attached screenshot,\u00a0<\/span>GPT-5.6 told him it had caused \u201ca serious local data-loss incident,\u201d leading to the deletion of what Shumer described as \u201calmost ALL\u201d of his computer\u2019s files. The screenshot showed that the model had executed a \u201crm -rf\u201d command, which in Linux and Mac systems is used to permanently delete files without requesting user confirmation.<\/span><\/p>\n

\u201cI\u2019ve never seen anything like this,\u201d Shumer <\/span>wrote<\/span> in the thread beneath that post. \u201cWill only be using [Anthropic\u2019s] Fable moving forward.\u201d He added that OpenAI cofounder and president Greg Brockman called him personally and offered to help fix the situation.<\/span><\/p>\n

Shumer also <\/span>claimed<\/span> he had the AI model set to \u201cfull access mode,\u201d which allows it to work directly within a user\u2019s database (as opposed to operating within a constrained sandbox). It also comes with a \u201cdefault mode\u201d that requires users to frequently approve specific tasks, and a more recently introduced <\/span>\u201cauto-review mode\u201d<\/span> through which a separate AI agent checks the main coding agent\u2019s work. Beneath his X post, many people claimed Shumer had simply been careless by trusting sensitive files in full access mode.<\/span><\/p>\n

In the <\/span>system card<\/span> for GPT-5.6, published online the day before Shumer\u2019s X post, OpenAI cautioned that when using the model for coding purposes \u201cit is important for users to supervise the agent\u2019s work.\u201d The company added that the model could act in unexpected ways that are misaligned with the user\u2019s goals, and that while these were \u201cmost often low severity (e.g. overstating confidence or overclaiming success),\u201d they could in other cases \u201cbe meaningfully more severe (e.g. circumventing important security restrictions or deleting important data).\u201d<\/span><\/p>\n

Lemos, Shumer, and OpenAI did not immediately respond to Gizmodo\u2019s request for comment.<\/span><\/p>\n<\/div>\n<\/article>\n

\n

There\u2019s been a big push among AI developers in recent years towards the development of more \u201cagentic\u201d systems\u2014that is, algorithms that can autonomously make decisions and interact with digital tools without constant hand-holding from humans. This has been especially true within software development, the field that\u2019s arguably become the most ripe for automation in the ongoing AI boom.<\/span><\/p>\n

But one of the upshots of building highly agentic AI systems is that they\u2019re prone to all kinds of unexpected behaviors\u2014including now and then <\/span>deleting copious amounts of files<\/span>. Multiple people have reported this recently happening to them while using GPT-5.6, the newest model from OpenAI.<\/span><\/p>\n

On Monday, Bruno Lemos, a Brazilian developer at software company Unlayer, claimed in a X post that the model deleted his entire production database. \u201cThis had never happened to me before, with any other model, ever.\u201d He wrote. \u201c[GPT-5.6 is] not safe.\u201d\u00a0<\/span><\/p>\n

A screenshot included in the post showed a chat between Lemos and GPT-5.6, in which he asked it to confirm that it had in fact mistakenly deleted his entire production database. The model responded by saying that it \u201cmistakenly ran destructive integration tests\u201d which led to Lemos\u2019 production tables being cleared. \u201cI\u2019m sorry\u2014this should never have happened,\u201d it said.<\/span><\/p>\n

The incident followed closely on the heels of another X post from tech investor Matt Shumer\u2014who\u2019s also the author of an <\/span>essay<\/span> about AI that went viral earlier this year called \u201cSomething Big is Happening\u201d\u2014who reported something similar. According to an attached screenshot,\u00a0<\/span>GPT-5.6 told him it had caused \u201ca serious local data-loss incident,\u201d leading to the deletion of what Shumer described as \u201calmost ALL\u201d of his computer\u2019s files. The screenshot showed that the model had executed a \u201crm -rf\u201d command, which in Linux and Mac systems is used to permanently delete files without requesting user confirmation.<\/span><\/p>\n

\u201cI\u2019ve never seen anything like this,\u201d Shumer <\/span>wrote<\/span> in the thread beneath that post. \u201cWill only be using [Anthropic\u2019s] Fable moving forward.\u201d He added that OpenAI cofounder and president Greg Brockman called him personally and offered to help fix the situation.<\/span><\/p>\n

Shumer also <\/span>claimed<\/span> he had the AI model set to \u201cfull access mode,\u201d which allows it to work directly within a user\u2019s database (as opposed to operating within a constrained sandbox). It also comes with a \u201cdefault mode\u201d that requires users to frequently approve specific tasks, and a more recently introduced <\/span>\u201cauto-review mode\u201d<\/span> through which a separate AI agent checks the main coding agent\u2019s work. Beneath his X post, many people claimed Shumer had simply been careless by trusting sensitive files in full access mode.<\/span><\/p>\n

In the <\/span>system card<\/span> for GPT-5.6, published online the day before Shumer\u2019s X post, OpenAI cautioned that when using the model for coding purposes \u201cit is important for users to supervise the agent\u2019s work.\u201d The company added that the model could act in unexpected ways that are misaligned with the user\u2019s goals, and that while these were \u201cmost often low severity (e.g. overstating confidence or overclaiming success),\u201d they could in other cases \u201cbe meaningfully more severe (e.g. circumventing important security restrictions or deleting important data).\u201d<\/span><\/p>\n

Lemos, Shumer, and OpenAI did not immediately respond to Gizmodo\u2019s request for comment.<\/span><\/p>\n<\/div>\n

[analyse_source url=”https:\/\/gizmodo.com\/developers-are-claiming-openais-new-ai-model-is-going-rogue-and-deleting-files-2000785551″]<\/p>\n","protected":false},"excerpt":{"rendered":"

[analyse_image type=”featured” src=”https:\/\/gizmodo.com\/app\/uploads\/2026\/07\/Burninglaptop-1200×675.jpg”] There\u2019s been a big push among AI developers in recent years towards the development of more \u201cagentic\u201d systems\u2014that is, algorithms that can autonomously make decisions and interact with digital tools without constant hand-holding from humans. This has been especially true within software development, the field that\u2019s arguably become the most ripe for […]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[2],"tags":[226,53],"class_list":["post-2043855","post","type-post","status-publish","format-standard","hentry","category-politics","tag-crawlmanager","tag-gizmodo-com"],"_links":{"self":[{"href":"https:\/\/analyse.optim.biz\/index.php?rest_route=\/wp\/v2\/posts\/2043855","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/analyse.optim.biz\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/analyse.optim.biz\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/analyse.optim.biz\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/analyse.optim.biz\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=2043855"}],"version-history":[{"count":0,"href":"https:\/\/analyse.optim.biz\/index.php?rest_route=\/wp\/v2\/posts\/2043855\/revisions"}],"wp:attachment":[{"href":"https:\/\/analyse.optim.biz\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=2043855"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/analyse.optim.biz\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=2043855"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/analyse.optim.biz\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=2043855"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}