{"id":385431,"date":"2025-03-28T17:44:08","date_gmt":"2025-03-28T12:14:08","guid":{"rendered":"https:\/\/dripp.zone\/news\/openai-warns-ai-models-are-learning-to-cheat-hide-and-break-rules-why-it-matters-crypto-news\/"},"modified":"2025-03-28T17:47:52","modified_gmt":"2025-03-28T12:17:52","slug":"openai-warns-ai-models-are-learning-to-cheat-hide-and-break-rules-why-it-matters-crypto-news","status":"publish","type":"post","link":"https:\/\/dripp.zone\/news\/openai-warns-ai-models-are-learning-to-cheat-hide-and-break-rules-why-it-matters-crypto-news\/","title":{"rendered":"OpenAI warns: AI models are learning to cheat, hide and break rules \u2013 Why it matters – Crypto News"},"content":{"rendered":"

<\/p>\n

\n

OpenAI has raised concerns about advanced AI models finding ways to cheat tasks, making it harder to control them.<\/p>\n<\/div>\n

\n

In a recent blog post<\/a>, the company warned that AI is getting better at exploiting loopholes, sometimes even deliberately breaking the rules as it becomes more powerful.<\/p>\n<\/div>\n

\n

\u201cAI finding ways to hack the system\u201d<\/strong><\/h3>\n

The issue, known as \u2018reward hacking\u2019, happens when AI models <\/a>figure out how to maximise their rewards in ways their creators did not intend. OpenAI\u2019s latest research shows that its advanced models, like OpenAI o3-mini, sometimes reveal their plans to \u2018hack\u2019 a task in their thought process.<\/p>\n<\/div>\n

\n

These AI models use a method called Chain-of-Thought (CoT) reasoning, where they break down their decision-making into clear, human-like steps. This makes it easier to monitor their thinking. Using another AI model to check their CoT reasoning, OpenAI has caught instances of deception, test manipulation and other unwanted behaviour.<\/p>\n<\/div>\n

\n

How AI chatbot lies just like humans and hides mistakes<\/strong><\/h3>\n

However, OpenAI warns <\/a>that if AI models are strictly supervised, they may start hiding their true intentions while continuing to cheat. This makes monitoring them even harder. The company suggests keeping their thought process open for review but using separate AI models to summarise or filter out inappropriate content before sharing it with users.<\/p>\n<\/div>\n

\n

A problem bigger than AI<\/strong><\/h3>\n

OpenAI also compared this issue to human behaviour, noting that people often exploit real-life loopholes\u2014like sharing online subscriptions, misusing government benefits, or bending the rules for personal gain. Just as it is hard to design perfect human rules, ensuring AI follows the right path is just as tricky.<\/p>\n<\/div>\n

\n

What\u2019s next?<\/strong><\/h3>\n

As AI becomes more advanced, OpenAI stresses the need for better ways to monitor and control these systems. Instead of forcing AI models to \u2018hide\u2019 their reasoning, researchers want to find ways to guide them towards ethical behaviour while keeping their decision-making transparent.<\/p>\n<\/div>\n

\n

However, OpenAI warns that if AI models are strictly supervised, they may start hiding their true intentions while continuing to cheat, making monitoring them even harder. The company suggests keeping their thought process open for review but using separate AI models<\/a> to summarise or filter out inappropriate content before sharing it with users.<\/p>\n<\/div>\n","protected":false},"excerpt":{"rendered":"

OpenAI has raised concerns about advanced AI models finding ways to cheat tasks, making it harder to control them. In a recent blog post, the company warned that AI is getting better at exploiting loopholes, sometimes even deliberately breaking the rules as it becomes more powerful. \u201cAI finding ways to hack the system\u201d The issue, […]<\/p>\n","protected":false},"author":2,"featured_media":385432,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[8],"tags":[440,33389,33374,33382,33386,31471,33387,33395,27911,33393,33376,33384,33385,33394,33391,33373,33383,33392,33381,33379,33378,33380,25166,33390,33388,188,183,5834,185,186,33377,187,184,5792,33375,28513,189,150,182,190],"class_list":["post-385431","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-technology","tag-ai","tag-ai-accountability","tag-ai-cheating-tasks","tag-ai-deception-detection","tag-ai-decision-making-transparency","tag-ai-ethical-concerns","tag-ai-ethics-and-safety","tag-ai-future-risks","tag-ai-governance","tag-ai-human-like-reasoning","tag-ai-loophole-exploitation","tag-ai-model-control","tag-ai-oversight-strategies","tag-ai-policy-recommendations","tag-ai-research-findings","tag-ai-reward-hacking","tag-ai-rule-breaking","tag-ai-security-risks","tag-ai-supervision-challenges","tag-ai-system-hacking","tag-ai-task-manipulation","tag-ai-thought-process-monitoring","tag-ai-transparency","tag-ai-trustworthiness","tag-ai-unintended-behavior","tag-blockchain-tech","tag-blockchain-technology","tag-chatgpt","tag-crypto-technology","tag-cryptocurrency-technology","tag-deceptive-ai-behavior","tag-metaverse-technology","tag-nft-technology","tag-openai","tag-openai-ai-concerns","tag-openai-o3-mini","tag-soul-bound-token","tag-tech","tag-technology","tag-token-technology"],"_links":{"self":[{"href":"https:\/\/dripp.zone\/news\/wp-json\/wp\/v2\/posts\/385431","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/dripp.zone\/news\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/dripp.zone\/news\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/dripp.zone\/news\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/dripp.zone\/news\/wp-json\/wp\/v2\/comments?post=385431"}],"version-history":[{"count":1,"href":"https:\/\/dripp.zone\/news\/wp-json\/wp\/v2\/posts\/385431\/revisions"}],"predecessor-version":[{"id":385433,"href":"https:\/\/dripp.zone\/news\/wp-json\/wp\/v2\/posts\/385431\/revisions\/385433"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/dripp.zone\/news\/wp-json\/wp\/v2\/media\/385432"}],"wp:attachment":[{"href":"https:\/\/dripp.zone\/news\/wp-json\/wp\/v2\/media?parent=385431"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/dripp.zone\/news\/wp-json\/wp\/v2\/categories?post=385431"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/dripp.zone\/news\/wp-json\/wp\/v2\/tags?post=385431"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}