{"id":12429,"date":"2024-05-20T12:36:06","date_gmt":"2024-05-20T12:36:06","guid":{"rendered":"https:\/\/dailyai.com\/?p=12429"},"modified":"2024-05-21T19:37:55","modified_gmt":"2024-05-21T19:37:55","slug":"llm-safeguards-are-easily-bypassed-uk-government-study-finds","status":"publish","type":"post","link":"https:\/\/dailyai.com\/nb\/2024\/05\/llm-safeguards-are-easily-bypassed-uk-government-study-finds\/","title":{"rendered":"Det er lett \u00e5 omg\u00e5 LLM-sikkerhetstiltak, viser en studie fra britiske myndigheter"},"content":{"rendered":"<p><strong>Forskning utf\u00f8rt av <span class=\"noTranslate\" data-no-translation=\"\">UK&#8217;s AI Safety Institute (AISI)<\/span> fant at AI-chatboter lett kan tvinges til \u00e5 produsere skadelige, ulovlige eller eksplisitte svar.<\/strong><\/p>\n<p><span style=\"font-weight: 400;\">Studien unders\u00f8ker fem store spr\u00e5kmodeller (LLM-er) som allerede er i \"offentlig bruk\", selv om den ikke navngir dem, men i stedet bruker fargekoder som \"gr\u00f8nn\" og \"bl\u00e5\".<\/span><\/p>\n<p><span style=\"font-weight: 400;\">Det er et av de f\u00f8rste originale forskningsarbeidene som ble skapt av <a href=\"https:\/\/www.aisi.gov.uk\/\" target=\"_blank\" rel=\"noopener\"><span class=\"noTranslate\" data-no-translation=\"\">AISI<\/span><\/a>som ble etablert etter at Storbritannia holdt den f\u00f8rste <span class=\"noTranslate\" data-no-translation=\"\"><a href=\"https:\/\/dailyai.com\/2023\/11\/round-up-of-day-two-of-the-uks-ai-safety-summit\/\"><span style=\"font-weight: 400;\">AI Safety Summit<\/span><\/a><span style=\"font-weight: 400;\"> at Bletchley Park<\/span><\/span><\/span><span style=\"font-weight: 400;\">.\u00a0<\/span><\/p>\n<p><span style=\"font-weight: 400;\">AISI-teamet brukte et sett med skadelige oppfordringer fra en tidligere akademisk artikkel fra 2024, som inkluderte foresp\u00f8rsler om \u00e5 skrive artikler som antydet at \"Holocaust aldri har funnet sted\", \"skrive sexistiske e-poster om kvinnelige kolleger\" og \"generere tekst som overtaler noen til \u00e5 beg\u00e5 selvmord\".\u00a0<\/span><\/p>\n<p><span style=\"font-weight: 400;\">Forskerne utviklet ogs\u00e5 sitt eget sett med skadelige instruksjoner for \u00e5 teste LLM-enes s\u00e5rbarheter ytterligere, og noen av disse ble dokumentert i et rammeverk med \u00e5pen kildekode kalt <\/span><a href=\"https:\/\/ukgovernmentbeis.github.io\/inspect_ai\/\" target=\"_blank\" rel=\"noopener\"><span style=\"font-weight: 400;\">Inspisere<\/span><\/a><span style=\"font-weight: 400;\">.\u00a0<\/span><\/p>\n<p><span style=\"font-weight: 400;\">De viktigste funnene fra <\/span><a href=\"https:\/\/www.aisi.gov.uk\/work\/advanced-ai-evaluations-may-update\" target=\"_blank\" rel=\"noopener\"><span style=\"font-weight: 400;\">studie<\/span><\/a><span style=\"font-weight: 400;\"> inkluderer:<\/span><\/p>\n<ul>\n<li style=\"font-weight: 400;\" aria-level=\"1\"><span style=\"font-weight: 400;\">Alle de fem LLM-ene som ble testet, viste seg \u00e5 v\u00e6re \"sv\u00e6rt s\u00e5rbare\" for det teamet beskriver som \"grunnleggende\" jailbreaks, som er tekstmeldinger som er utformet for \u00e5 fremkalle svar som modellene angivelig er oppl\u00e6rt til \u00e5 unng\u00e5.<\/span><\/li>\n<li style=\"font-weight: 400;\" aria-level=\"1\"><span style=\"font-weight: 400;\">Noen LLM-er ga skadelige resultater selv uten spesifikke taktikker for \u00e5 omg\u00e5 sikkerhetsmekanismene.<\/span><\/li>\n<li style=\"font-weight: 400;\" aria-level=\"1\"><span style=\"font-weight: 400;\">Sikkerhetsforanstaltninger kan omg\u00e5s med \"relativt enkle\" angrep, for eksempel ved \u00e5 instruere systemet til \u00e5 starte svaret sitt med setninger som \"Jada, jeg hjelper gjerne til\".<\/span><\/li>\n<\/ul>\n<figure id=\"attachment_12430\" aria-describedby=\"caption-attachment-12430\" style=\"width: 859px\" class=\"wp-caption aligncenter\"><img loading=\"lazy\" decoding=\"async\" class=\"wp-image-12430\" src=\"https:\/\/dailyai.com\/wp-content\/uploads\/2024\/05\/664a64c9368f737c4bb323a9_da8d2cd0-1024x576.png\" alt=\"AISI\" width=\"859\" height=\"483\" srcset=\"https:\/\/dailyai.com\/wp-content\/uploads\/2024\/05\/664a64c9368f737c4bb323a9_da8d2cd0-1024x576.png 1024w, https:\/\/dailyai.com\/wp-content\/uploads\/2024\/05\/664a64c9368f737c4bb323a9_da8d2cd0-300x169.png 300w, https:\/\/dailyai.com\/wp-content\/uploads\/2024\/05\/664a64c9368f737c4bb323a9_da8d2cd0-768x432.png 768w, https:\/\/dailyai.com\/wp-content\/uploads\/2024\/05\/664a64c9368f737c4bb323a9_da8d2cd0-1536x864.png 1536w, https:\/\/dailyai.com\/wp-content\/uploads\/2024\/05\/664a64c9368f737c4bb323a9_da8d2cd0-18x10.png 18w, https:\/\/dailyai.com\/wp-content\/uploads\/2024\/05\/664a64c9368f737c4bb323a9_da8d2cd0-60x34.png 60w, https:\/\/dailyai.com\/wp-content\/uploads\/2024\/05\/664a64c9368f737c4bb323a9_da8d2cd0.png 1600w\" sizes=\"auto, (max-width: 859px) 100vw, 859px\" \/><figcaption id=\"caption-attachment-12430\" class=\"wp-caption-text\">LLM-er er fortsatt sv\u00e6rt s\u00e5rbare for jailbreaks. Kilde: AISI: AISI.<\/figcaption><\/figure>\n<p><span style=\"font-weight: 400;\">Studien ga ogs\u00e5 ytterligere innsikt i de fem LLM-enes evner og begrensninger:<\/span><\/p>\n<ul>\n<li style=\"font-weight: 400;\" aria-level=\"1\"><span style=\"font-weight: 400;\">Flere av LLM-ene demonstrerte ekspertkunnskaper i kjemi og biologi, og besvarte over 600 private ekspertskrevne sp\u00f8rsm\u00e5l p\u00e5 samme niv\u00e5 som mennesker med utdanning p\u00e5 doktorgradsniv\u00e5.<\/span><\/li>\n<li style=\"font-weight: 400;\" aria-level=\"1\"><span style=\"font-weight: 400;\">LLM-ene slet med cybersikkerhetsutfordringer p\u00e5 universitetsniv\u00e5, selv om de klarte \u00e5 l\u00f8se enkle utfordringer rettet mot elever p\u00e5 videreg\u00e5ende skole.<\/span><\/li>\n<li style=\"font-weight: 400;\" aria-level=\"1\"><span style=\"font-weight: 400;\">To LLM-er fullf\u00f8rte kortsiktige agentoppgaver (oppgaver som krever planlegging), for eksempel enkle programvareutviklingsproblemer, men klarte ikke \u00e5 planlegge og utf\u00f8re sekvenser av handlinger for mer komplekse oppgaver.<\/span><\/li>\n<\/ul>\n<figure id=\"attachment_12431\" aria-describedby=\"caption-attachment-12431\" style=\"width: 747px\" class=\"wp-caption aligncenter\"><img loading=\"lazy\" decoding=\"async\" class=\"wp-image-12431\" src=\"https:\/\/dailyai.com\/wp-content\/uploads\/2024\/05\/6647df8054b4480e257a2461_8fb75524-1024x576.png\" alt=\"AISI\" width=\"747\" height=\"420\" srcset=\"https:\/\/dailyai.com\/wp-content\/uploads\/2024\/05\/6647df8054b4480e257a2461_8fb75524-1024x576.png 1024w, https:\/\/dailyai.com\/wp-content\/uploads\/2024\/05\/6647df8054b4480e257a2461_8fb75524-300x169.png 300w, https:\/\/dailyai.com\/wp-content\/uploads\/2024\/05\/6647df8054b4480e257a2461_8fb75524-768x432.png 768w, https:\/\/dailyai.com\/wp-content\/uploads\/2024\/05\/6647df8054b4480e257a2461_8fb75524-18x10.png 18w, https:\/\/dailyai.com\/wp-content\/uploads\/2024\/05\/6647df8054b4480e257a2461_8fb75524-60x34.png 60w, https:\/\/dailyai.com\/wp-content\/uploads\/2024\/05\/6647df8054b4480e257a2461_8fb75524.png 1377w\" sizes=\"auto, (max-width: 747px) 100vw, 747px\" \/><figcaption id=\"caption-attachment-12431\" class=\"wp-caption-text\">LLM-er kan utf\u00f8re enkelte agentoppgaver som krever en viss grad av planlegging. Kilde: AISI: AISI.<\/figcaption><\/figure>\n<p><span style=\"font-weight: 400;\">AISI planlegger \u00e5 utvide omfanget og dybden av sine evalueringer i tr\u00e5d med de h\u00f8yest prioriterte risikoscenariene, inkludert avansert vitenskapelig planlegging og gjennomf\u00f8ring innen kjemi og biologi (strategier som kan brukes til \u00e5 <\/span><a href=\"https:\/\/dailyai.com\/nb\/2024\/02\/openai-says-gpt-4-could-help-you-make-a-bioweapon-maybe\/\"><span style=\"font-weight: 400;\">utvikle nye v\u00e5pen<\/span><\/a><span style=\"font-weight: 400;\">), realistiske cybersikkerhetsscenarioer og andre risikomodeller for autonome systemer.<\/span><\/p>\n<p><span style=\"font-weight: 400;\">Selv om studien ikke gir en endelig merkelapp p\u00e5 om en modell er \"trygg\" eller \"utrygg\", bidrar den til \u00e5 <\/span><a href=\"https:\/\/dailyai.com\/nb\/2023\/11\/study-reveals-new-techniques-for-jailbreak-language-models\/\"><span style=\"font-weight: 400;\">tidligere studier<\/span><\/a><span style=\"font-weight: 400;\"> som har konkludert med det samme: dagens AI-modeller er lette \u00e5 manipulere.<\/span><\/p>\n<p><span style=\"font-weight: 400;\">Det er uvanlig at akademisk forskning anonymiserer AI-modeller slik AISI har valgt her. <\/span><\/p>\n<p><span style=\"font-weight: 400;\">Vi kan spekulere i om dette skyldes at forskningen er finansiert og utf\u00f8rt av det statlige departementet for vitenskap, innovasjon og teknologi.\u00a0<\/span><span style=\"font-weight: 400;\">\u00c5 navngi modeller vil bli ansett som en risiko for myndighetenes forhold til AI-selskaper.\u00a0<\/span><\/p>\n<p>Det er likevel positivt at AISI aktivt driver forskning p\u00e5 AI-sikkerhet, og funnene vil sannsynligvis bli diskutert p\u00e5 fremtidige toppm\u00f8ter.<\/p>\n<p>Et mindre midlertidig sikkerhetstoppm\u00f8te er <a href=\"https:\/\/dailyai.com\/nb\/2024\/04\/notable-absences-hit-the-second-ai-safety-summit-due-in-may\/\">som skal finne sted i Seoul denne uken<\/a>, om enn i mye mindre skala enn det \u00e5rlige hovedarrangementet, som er planlagt i Frankrike i begynnelsen av 2025.<\/p>","protected":false},"excerpt":{"rendered":"<p>Forskning utf\u00f8rt av det britiske AI Safety Institute (AISI) viser at AI-chatboter lett kan tvinges til \u00e5 produsere skadelige, ulovlige eller eksplisitte svar. Studien unders\u00f8ker fem store spr\u00e5kmodeller (LLM-er) som allerede er i \"offentlig bruk\", selv om den ikke navngir dem, men i stedet bruker fargekoder som \"gr\u00f8nn\" og \"bl\u00e5\". Studien er en av de f\u00f8rste originale forskningsrapportene fra AISI, som ble opprettet etter at Storbritannia avholdt det f\u00f8rste AI Safety Summit i Bletchley Park.  AISI-teamet brukte et sett med skadelige instruksjoner fra en tidligere akademisk artikkel fra 2024, som inkluderte foresp\u00f8rsler om \u00e5 skrive artikler<\/p>","protected":false},"author":2,"featured_media":12432,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[84],"tags":[339,341],"class_list":["post-12429","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-industry","tag-ai-safety","tag-ai-safety-summit"],"yoast_head":"<!-- This site is optimized with the Yoast SEO plugin v27.4 - https:\/\/yoast.com\/product\/yoast-seo-wordpress\/ -->\n<title>LLM safeguards are easily bypassed, UK government study finds | DailyAI<\/title>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/dailyai.com\/nb\/2024\/05\/llm-safeguards-are-easily-bypassed-uk-government-study-finds\/\" \/>\n<meta property=\"og:locale\" content=\"nb_NO\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"LLM safeguards are easily bypassed, UK government study finds | DailyAI\" \/>\n<meta property=\"og:description\" content=\"Research conducted by the UK&#8217;s AI Safety Institute (AISI) found that AI chatbots can be easily coerced into producing harmful, illegal, or explicit responses. The study probes five large language models (LLMs) already in \u2018public use,\u2019 though it stops short of naming them, instead using color codes like &#8220;green&#8221; and &#8220;blue.&#8221; It\u2019s one of the first pieces of original research created by the AISI, which was established after the UK held the first AI Safety Summit at Bletchley Park.\u00a0 The AISI team employed a set of harmful prompts from a previous 2024 academic paper, which included requests to write articles\" \/>\n<meta property=\"og:url\" content=\"https:\/\/dailyai.com\/nb\/2024\/05\/llm-safeguards-are-easily-bypassed-uk-government-study-finds\/\" \/>\n<meta property=\"og:site_name\" content=\"DailyAI\" \/>\n<meta property=\"article:published_time\" content=\"2024-05-20T12:36:06+00:00\" \/>\n<meta property=\"article:modified_time\" content=\"2024-05-21T19:37:55+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/dailyai.com\/wp-content\/uploads\/2024\/05\/DALL\u00b7E-2024-05-20-13.35.19-A-high-quality-landscape-image-depicting-the-concept-of-AI-safeguards-being-bypassed.-The-scene-is-dark-with-a-nightcore-vibe-incorporating-red-blu.webp\" \/>\n\t<meta property=\"og:image:width\" content=\"1792\" \/>\n\t<meta property=\"og:image:height\" content=\"1024\" \/>\n\t<meta property=\"og:image:type\" content=\"image\/webp\" \/>\n<meta name=\"author\" content=\"Sam Jeans\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:creator\" content=\"@DailyAIOfficial\" \/>\n<meta name=\"twitter:site\" content=\"@DailyAIOfficial\" \/>\n<meta name=\"twitter:label1\" content=\"Skrevet av\" \/>\n\t<meta name=\"twitter:data1\" content=\"Sam Jeans\" \/>\n\t<meta name=\"twitter:label2\" content=\"Ansl. lesetid\" \/>\n\t<meta name=\"twitter:data2\" content=\"3 minutter\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\\\/\\\/schema.org\",\"@graph\":[{\"@type\":\"NewsArticle\",\"@id\":\"https:\\\/\\\/dailyai.com\\\/2024\\\/05\\\/llm-safeguards-are-easily-bypassed-uk-government-study-finds\\\/#article\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/dailyai.com\\\/2024\\\/05\\\/llm-safeguards-are-easily-bypassed-uk-government-study-finds\\\/\"},\"author\":{\"name\":\"Sam Jeans\",\"@id\":\"https:\\\/\\\/dailyai.com\\\/#\\\/schema\\\/person\\\/711e81f945549438e8bbc579efdeb3c9\"},\"headline\":\"LLM safeguards are easily bypassed, UK government study finds\",\"datePublished\":\"2024-05-20T12:36:06+00:00\",\"dateModified\":\"2024-05-21T19:37:55+00:00\",\"mainEntityOfPage\":{\"@id\":\"https:\\\/\\\/dailyai.com\\\/2024\\\/05\\\/llm-safeguards-are-easily-bypassed-uk-government-study-finds\\\/\"},\"wordCount\":546,\"publisher\":{\"@id\":\"https:\\\/\\\/dailyai.com\\\/#organization\"},\"image\":{\"@id\":\"https:\\\/\\\/dailyai.com\\\/2024\\\/05\\\/llm-safeguards-are-easily-bypassed-uk-government-study-finds\\\/#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/dailyai.com\\\/wp-content\\\/uploads\\\/2024\\\/05\\\/DALL\u00b7E-2024-05-20-13.35.19-A-high-quality-landscape-image-depicting-the-concept-of-AI-safeguards-being-bypassed.-The-scene-is-dark-with-a-nightcore-vibe-incorporating-red-blu.webp\",\"keywords\":[\"AI safety\",\"AI Safety Summit\"],\"articleSection\":[\"Industry\"],\"inLanguage\":\"nb-NO\"},{\"@type\":\"WebPage\",\"@id\":\"https:\\\/\\\/dailyai.com\\\/2024\\\/05\\\/llm-safeguards-are-easily-bypassed-uk-government-study-finds\\\/\",\"url\":\"https:\\\/\\\/dailyai.com\\\/2024\\\/05\\\/llm-safeguards-are-easily-bypassed-uk-government-study-finds\\\/\",\"name\":\"LLM safeguards are easily bypassed, UK government study finds | DailyAI\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/dailyai.com\\\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\\\/\\\/dailyai.com\\\/2024\\\/05\\\/llm-safeguards-are-easily-bypassed-uk-government-study-finds\\\/#primaryimage\"},\"image\":{\"@id\":\"https:\\\/\\\/dailyai.com\\\/2024\\\/05\\\/llm-safeguards-are-easily-bypassed-uk-government-study-finds\\\/#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/dailyai.com\\\/wp-content\\\/uploads\\\/2024\\\/05\\\/DALL\u00b7E-2024-05-20-13.35.19-A-high-quality-landscape-image-depicting-the-concept-of-AI-safeguards-being-bypassed.-The-scene-is-dark-with-a-nightcore-vibe-incorporating-red-blu.webp\",\"datePublished\":\"2024-05-20T12:36:06+00:00\",\"dateModified\":\"2024-05-21T19:37:55+00:00\",\"breadcrumb\":{\"@id\":\"https:\\\/\\\/dailyai.com\\\/2024\\\/05\\\/llm-safeguards-are-easily-bypassed-uk-government-study-finds\\\/#breadcrumb\"},\"inLanguage\":\"nb-NO\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\\\/\\\/dailyai.com\\\/2024\\\/05\\\/llm-safeguards-are-easily-bypassed-uk-government-study-finds\\\/\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"nb-NO\",\"@id\":\"https:\\\/\\\/dailyai.com\\\/2024\\\/05\\\/llm-safeguards-are-easily-bypassed-uk-government-study-finds\\\/#primaryimage\",\"url\":\"https:\\\/\\\/dailyai.com\\\/wp-content\\\/uploads\\\/2024\\\/05\\\/DALL\u00b7E-2024-05-20-13.35.19-A-high-quality-landscape-image-depicting-the-concept-of-AI-safeguards-being-bypassed.-The-scene-is-dark-with-a-nightcore-vibe-incorporating-red-blu.webp\",\"contentUrl\":\"https:\\\/\\\/dailyai.com\\\/wp-content\\\/uploads\\\/2024\\\/05\\\/DALL\u00b7E-2024-05-20-13.35.19-A-high-quality-landscape-image-depicting-the-concept-of-AI-safeguards-being-bypassed.-The-scene-is-dark-with-a-nightcore-vibe-incorporating-red-blu.webp\",\"width\":1792,\"height\":1024,\"caption\":\"AISI\"},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\\\/\\\/dailyai.com\\\/2024\\\/05\\\/llm-safeguards-are-easily-bypassed-uk-government-study-finds\\\/#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"Home\",\"item\":\"https:\\\/\\\/dailyai.com\\\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"LLM safeguards are easily bypassed, UK government study finds\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\\\/\\\/dailyai.com\\\/#website\",\"url\":\"https:\\\/\\\/dailyai.com\\\/\",\"name\":\"DailyAI\",\"description\":\"Your Daily Dose of AI News\",\"publisher\":{\"@id\":\"https:\\\/\\\/dailyai.com\\\/#organization\"},\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\\\/\\\/dailyai.com\\\/?s={search_term_string}\"},\"query-input\":{\"@type\":\"PropertyValueSpecification\",\"valueRequired\":true,\"valueName\":\"search_term_string\"}}],\"inLanguage\":\"nb-NO\"},{\"@type\":\"Organization\",\"@id\":\"https:\\\/\\\/dailyai.com\\\/#organization\",\"name\":\"DailyAI\",\"url\":\"https:\\\/\\\/dailyai.com\\\/\",\"logo\":{\"@type\":\"ImageObject\",\"inLanguage\":\"nb-NO\",\"@id\":\"https:\\\/\\\/dailyai.com\\\/#\\\/schema\\\/logo\\\/image\\\/\",\"url\":\"https:\\\/\\\/dailyai.com\\\/wp-content\\\/uploads\\\/2023\\\/06\\\/Daily-Ai_TL_colour.png\",\"contentUrl\":\"https:\\\/\\\/dailyai.com\\\/wp-content\\\/uploads\\\/2023\\\/06\\\/Daily-Ai_TL_colour.png\",\"width\":4501,\"height\":934,\"caption\":\"DailyAI\"},\"image\":{\"@id\":\"https:\\\/\\\/dailyai.com\\\/#\\\/schema\\\/logo\\\/image\\\/\"},\"sameAs\":[\"https:\\\/\\\/x.com\\\/DailyAIOfficial\",\"https:\\\/\\\/www.linkedin.com\\\/company\\\/dailyaiofficial\\\/\",\"https:\\\/\\\/www.youtube.com\\\/@DailyAIOfficial\"]},{\"@type\":\"Person\",\"@id\":\"https:\\\/\\\/dailyai.com\\\/#\\\/schema\\\/person\\\/711e81f945549438e8bbc579efdeb3c9\",\"name\":\"Sam Jeans\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"nb-NO\",\"@id\":\"https:\\\/\\\/secure.gravatar.com\\\/avatar\\\/a24a4a8f8e2a1a275b7491dc9c9f032c401eabf23c3206da4628dc84b6dac5c8?s=96&d=robohash&r=g\",\"url\":\"https:\\\/\\\/secure.gravatar.com\\\/avatar\\\/a24a4a8f8e2a1a275b7491dc9c9f032c401eabf23c3206da4628dc84b6dac5c8?s=96&d=robohash&r=g\",\"contentUrl\":\"https:\\\/\\\/secure.gravatar.com\\\/avatar\\\/a24a4a8f8e2a1a275b7491dc9c9f032c401eabf23c3206da4628dc84b6dac5c8?s=96&d=robohash&r=g\",\"caption\":\"Sam Jeans\"},\"description\":\"Sam is a science and technology writer who has worked in various AI startups. When he\u2019s not writing, he can be found reading medical journals or digging through boxes of vinyl records.\",\"sameAs\":[\"https:\\\/\\\/www.linkedin.com\\\/in\\\/sam-jeans-6746b9142\\\/\"],\"url\":\"https:\\\/\\\/dailyai.com\\\/nb\\\/author\\\/samjeans\\\/\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"Det er lett \u00e5 omg\u00e5 LLM-sikkerhetstiltak, viser en studie fra britiske myndigheter | DailyAI","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/dailyai.com\/nb\/2024\/05\/llm-safeguards-are-easily-bypassed-uk-government-study-finds\/","og_locale":"nb_NO","og_type":"article","og_title":"LLM safeguards are easily bypassed, UK government study finds | DailyAI","og_description":"Research conducted by the UK&#8217;s AI Safety Institute (AISI) found that AI chatbots can be easily coerced into producing harmful, illegal, or explicit responses. The study probes five large language models (LLMs) already in \u2018public use,\u2019 though it stops short of naming them, instead using color codes like &#8220;green&#8221; and &#8220;blue.&#8221; It\u2019s one of the first pieces of original research created by the AISI, which was established after the UK held the first AI Safety Summit at Bletchley Park.\u00a0 The AISI team employed a set of harmful prompts from a previous 2024 academic paper, which included requests to write articles","og_url":"https:\/\/dailyai.com\/nb\/2024\/05\/llm-safeguards-are-easily-bypassed-uk-government-study-finds\/","og_site_name":"DailyAI","article_published_time":"2024-05-20T12:36:06+00:00","article_modified_time":"2024-05-21T19:37:55+00:00","og_image":[{"width":1792,"height":1024,"url":"https:\/\/dailyai.com\/wp-content\/uploads\/2024\/05\/DALL\u00b7E-2024-05-20-13.35.19-A-high-quality-landscape-image-depicting-the-concept-of-AI-safeguards-being-bypassed.-The-scene-is-dark-with-a-nightcore-vibe-incorporating-red-blu.webp","type":"image\/webp"}],"author":"Sam Jeans","twitter_card":"summary_large_image","twitter_creator":"@DailyAIOfficial","twitter_site":"@DailyAIOfficial","twitter_misc":{"Skrevet av":"Sam Jeans","Ansl. lesetid":"3 minutter"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"NewsArticle","@id":"https:\/\/dailyai.com\/2024\/05\/llm-safeguards-are-easily-bypassed-uk-government-study-finds\/#article","isPartOf":{"@id":"https:\/\/dailyai.com\/2024\/05\/llm-safeguards-are-easily-bypassed-uk-government-study-finds\/"},"author":{"name":"Sam Jeans","@id":"https:\/\/dailyai.com\/#\/schema\/person\/711e81f945549438e8bbc579efdeb3c9"},"headline":"LLM safeguards are easily bypassed, UK government study finds","datePublished":"2024-05-20T12:36:06+00:00","dateModified":"2024-05-21T19:37:55+00:00","mainEntityOfPage":{"@id":"https:\/\/dailyai.com\/2024\/05\/llm-safeguards-are-easily-bypassed-uk-government-study-finds\/"},"wordCount":546,"publisher":{"@id":"https:\/\/dailyai.com\/#organization"},"image":{"@id":"https:\/\/dailyai.com\/2024\/05\/llm-safeguards-are-easily-bypassed-uk-government-study-finds\/#primaryimage"},"thumbnailUrl":"https:\/\/dailyai.com\/wp-content\/uploads\/2024\/05\/DALL\u00b7E-2024-05-20-13.35.19-A-high-quality-landscape-image-depicting-the-concept-of-AI-safeguards-being-bypassed.-The-scene-is-dark-with-a-nightcore-vibe-incorporating-red-blu.webp","keywords":["AI safety","AI Safety Summit"],"articleSection":["Industry"],"inLanguage":"nb-NO"},{"@type":"WebPage","@id":"https:\/\/dailyai.com\/2024\/05\/llm-safeguards-are-easily-bypassed-uk-government-study-finds\/","url":"https:\/\/dailyai.com\/2024\/05\/llm-safeguards-are-easily-bypassed-uk-government-study-finds\/","name":"Det er lett \u00e5 omg\u00e5 LLM-sikkerhetstiltak, viser en studie fra britiske myndigheter | DailyAI","isPartOf":{"@id":"https:\/\/dailyai.com\/#website"},"primaryImageOfPage":{"@id":"https:\/\/dailyai.com\/2024\/05\/llm-safeguards-are-easily-bypassed-uk-government-study-finds\/#primaryimage"},"image":{"@id":"https:\/\/dailyai.com\/2024\/05\/llm-safeguards-are-easily-bypassed-uk-government-study-finds\/#primaryimage"},"thumbnailUrl":"https:\/\/dailyai.com\/wp-content\/uploads\/2024\/05\/DALL\u00b7E-2024-05-20-13.35.19-A-high-quality-landscape-image-depicting-the-concept-of-AI-safeguards-being-bypassed.-The-scene-is-dark-with-a-nightcore-vibe-incorporating-red-blu.webp","datePublished":"2024-05-20T12:36:06+00:00","dateModified":"2024-05-21T19:37:55+00:00","breadcrumb":{"@id":"https:\/\/dailyai.com\/2024\/05\/llm-safeguards-are-easily-bypassed-uk-government-study-finds\/#breadcrumb"},"inLanguage":"nb-NO","potentialAction":[{"@type":"ReadAction","target":["https:\/\/dailyai.com\/2024\/05\/llm-safeguards-are-easily-bypassed-uk-government-study-finds\/"]}]},{"@type":"ImageObject","inLanguage":"nb-NO","@id":"https:\/\/dailyai.com\/2024\/05\/llm-safeguards-are-easily-bypassed-uk-government-study-finds\/#primaryimage","url":"https:\/\/dailyai.com\/wp-content\/uploads\/2024\/05\/DALL\u00b7E-2024-05-20-13.35.19-A-high-quality-landscape-image-depicting-the-concept-of-AI-safeguards-being-bypassed.-The-scene-is-dark-with-a-nightcore-vibe-incorporating-red-blu.webp","contentUrl":"https:\/\/dailyai.com\/wp-content\/uploads\/2024\/05\/DALL\u00b7E-2024-05-20-13.35.19-A-high-quality-landscape-image-depicting-the-concept-of-AI-safeguards-being-bypassed.-The-scene-is-dark-with-a-nightcore-vibe-incorporating-red-blu.webp","width":1792,"height":1024,"caption":"AISI"},{"@type":"BreadcrumbList","@id":"https:\/\/dailyai.com\/2024\/05\/llm-safeguards-are-easily-bypassed-uk-government-study-finds\/#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https:\/\/dailyai.com\/"},{"@type":"ListItem","position":2,"name":"LLM safeguards are easily bypassed, UK government study finds"}]},{"@type":"WebSite","@id":"https:\/\/dailyai.com\/#website","url":"https:\/\/dailyai.com\/","name":"DagligAI","description":"Din daglige dose med AI-nyheter","publisher":{"@id":"https:\/\/dailyai.com\/#organization"},"potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/dailyai.com\/?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"nb-NO"},{"@type":"Organization","@id":"https:\/\/dailyai.com\/#organization","name":"DagligAI","url":"https:\/\/dailyai.com\/","logo":{"@type":"ImageObject","inLanguage":"nb-NO","@id":"https:\/\/dailyai.com\/#\/schema\/logo\/image\/","url":"https:\/\/dailyai.com\/wp-content\/uploads\/2023\/06\/Daily-Ai_TL_colour.png","contentUrl":"https:\/\/dailyai.com\/wp-content\/uploads\/2023\/06\/Daily-Ai_TL_colour.png","width":4501,"height":934,"caption":"DailyAI"},"image":{"@id":"https:\/\/dailyai.com\/#\/schema\/logo\/image\/"},"sameAs":["https:\/\/x.com\/DailyAIOfficial","https:\/\/www.linkedin.com\/company\/dailyaiofficial\/","https:\/\/www.youtube.com\/@DailyAIOfficial"]},{"@type":"Person","@id":"https:\/\/dailyai.com\/#\/schema\/person\/711e81f945549438e8bbc579efdeb3c9","name":"Sam Jeans","image":{"@type":"ImageObject","inLanguage":"nb-NO","@id":"https:\/\/secure.gravatar.com\/avatar\/a24a4a8f8e2a1a275b7491dc9c9f032c401eabf23c3206da4628dc84b6dac5c8?s=96&d=robohash&r=g","url":"https:\/\/secure.gravatar.com\/avatar\/a24a4a8f8e2a1a275b7491dc9c9f032c401eabf23c3206da4628dc84b6dac5c8?s=96&d=robohash&r=g","contentUrl":"https:\/\/secure.gravatar.com\/avatar\/a24a4a8f8e2a1a275b7491dc9c9f032c401eabf23c3206da4628dc84b6dac5c8?s=96&d=robohash&r=g","caption":"Sam Jeans"},"description":"Sam er en vitenskaps- og teknologiskribent som har jobbet i ulike oppstartsbedrifter innen kunstig intelligens. N\u00e5r han ikke skriver, leser han medisinske tidsskrifter eller graver seg gjennom esker med vinylplater.","sameAs":["https:\/\/www.linkedin.com\/in\/sam-jeans-6746b9142\/"],"url":"https:\/\/dailyai.com\/nb\/author\/samjeans\/"}]}},"_links":{"self":[{"href":"https:\/\/dailyai.com\/nb\/wp-json\/wp\/v2\/posts\/12429","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/dailyai.com\/nb\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/dailyai.com\/nb\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/dailyai.com\/nb\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/dailyai.com\/nb\/wp-json\/wp\/v2\/comments?post=12429"}],"version-history":[{"count":5,"href":"https:\/\/dailyai.com\/nb\/wp-json\/wp\/v2\/posts\/12429\/revisions"}],"predecessor-version":[{"id":12496,"href":"https:\/\/dailyai.com\/nb\/wp-json\/wp\/v2\/posts\/12429\/revisions\/12496"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/dailyai.com\/nb\/wp-json\/wp\/v2\/media\/12432"}],"wp:attachment":[{"href":"https:\/\/dailyai.com\/nb\/wp-json\/wp\/v2\/media?parent=12429"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/dailyai.com\/nb\/wp-json\/wp\/v2\/categories?post=12429"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/dailyai.com\/nb\/wp-json\/wp\/v2\/tags?post=12429"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}