{"id":277071,"date":"2026-09-19T02:00:27","date_gmt":"2026-09-19T02:00:27","guid":{"rendered":"https:\/\/ushopwell.com\/ublog\/llms-respond-differently-to-harmful-prompts-when-ai-watermarking-is-used\/"},"modified":"2026-09-19T02:00:27","modified_gmt":"2026-09-19T02:00:27","slug":"llms-respond-differently-to-harmful-prompts-when-ai-watermarking-is-used","status":"publish","type":"post","link":"https:\/\/ushopwell.com\/ublog\/llms-respond-differently-to-harmful-prompts-when-ai-watermarking-is-used\/","title":{"rendered":"LLMs respond differently to harmful prompts when AI watermarking is used"},"content":{"rendered":"
\n

In response to a new European Union law, AI platforms are implementing new schemes for watermarking the content they generate. Anthropic recently disclosed<\/a> its future Claude models will use SynthID-Text<\/a>, an approach Google created and released as open source. It uses a secret key that subtly changes the process a model uses for choosing the next word in a sentence. Whereas a top next word choice might be \u201ccloudy,\u201d the key might change it to \u201covercast.\u201d Anyone who knows the key can determine if it was generated by the platform using it.<\/p>\n

New research<\/a> shows that SynthID-Text can change not just word selection but also the tools a model invokes and the chances it will adhere to or disregard safety guardrails it has been trained to follow. The threat can become greater in the face of an adversarial prompt, in which an attacker attempts to cause a model to carry out a harmful action, such as revealing a password or other sensitive information. Instructions that normally wouldn\u2019t be followed will, in some cases, be performed once the watermarking is deployed. The finding underscores the need for developers to thoroughly test how their LLMs and agents behave when watermarking is in place.<\/p>\n

Changing safety behavior<\/h2>\n

\u201cAs compared to the same models without watermarking, it is definitely going to change their behavior, especially when we place it under adversarial conditions, or we make these models call tools when they\u2019re powering an agent,\u201d Andrea Siposova, an AI security researcher at Lasso Security, told Ars. \u201cWatermarking is made to not be perceptible to a reader, but we know that when we are changing anything about what the model is generating, it is going to cause some tradeoffs, it\u2019s going to show up somewhere.\u201d<\/p>\n

Read full article<\/a><\/p>\n

Comments<\/a><\/p>\n<\/div>\n

Powered by WPeMatico<\/a><\/small><\/p>\n","protected":false},"excerpt":{"rendered":"

In response to a new European Union law, AI platforms are implementing new schemes for watermarking the content they generate. Anthropic recently disclosed its future Claude models will use SynthID-Text, an approach Google created and released as open source. It uses a secret key that subtly changes the process a model uses for choosing the next word in a sentence. Whereas a top next word choice might be \u201ccloudy,\u201d the…<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":"","tve_updated_post":"","tve_custom_css":"","tve_user_custom_css":"","tve_globals":{},"tcb2_ready":0,"tcb_editor_enabled":0,"tve_landing_page":"","_tve_header":"","_tve_footer":""},"categories":[241],"tags":[],"class_list":["post-277071","post","type-post","status-publish","format-standard","hentry","category-technology"],"yoast_head":"\nLLMs respond differently to harmful prompts when AI watermarking is used - UshopWell.com<\/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:\/\/ushopwell.com\/ublog\/llms-respond-differently-to-harmful-prompts-when-ai-watermarking-is-used\/\" \/>\n<meta property=\"og:locale\" content=\"en_US\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"LLMs respond differently to harmful prompts when AI watermarking is used - UshopWell.com\" \/>\n<meta property=\"og:description\" content=\"In response to a new European Union law, AI platforms are implementing new schemes for watermarking the content they generate. Anthropic recently disclosed its future Claude models will use SynthID-Text, an approach Google created and released as open source. It uses a secret key that subtly changes the process a model uses for choosing the next word in a sentence. Whereas a top next word choice might be \u201ccloudy,\u201d the...\" \/>\n<meta property=\"og:url\" content=\"https:\/\/ushopwell.com\/ublog\/llms-respond-differently-to-harmful-prompts-when-ai-watermarking-is-used\/\" \/>\n<meta property=\"og:site_name\" content=\"UshopWell.com\" \/>\n<meta property=\"article:published_time\" content=\"2026-09-19T02:00:27+00:00\" \/>\n<meta name=\"author\" content=\"UShopWell\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:label1\" content=\"Written by\" \/>\n\t<meta name=\"twitter:data1\" content=\"UShopWell\" \/>\n\t<meta name=\"twitter:label2\" content=\"Est. reading time\" \/>\n\t<meta name=\"twitter:data2\" content=\"1 minute\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\\\/\\\/schema.org\",\"@graph\":[{\"@type\":\"Article\",\"@id\":\"https:\\\/\\\/ushopwell.com\\\/ublog\\\/llms-respond-differently-to-harmful-prompts-when-ai-watermarking-is-used\\\/#article\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/ushopwell.com\\\/ublog\\\/llms-respond-differently-to-harmful-prompts-when-ai-watermarking-is-used\\\/\"},\"author\":{\"name\":\"UShopWell\",\"@id\":\"https:\\\/\\\/ushopwell.com\\\/ublog\\\/#\\\/schema\\\/person\\\/6fd1f9e0ff932e534c86c70d5acff0fc\"},\"headline\":\"LLMs respond differently to harmful prompts when AI watermarking is used\",\"datePublished\":\"2026-09-19T02:00:27+00:00\",\"mainEntityOfPage\":{\"@id\":\"https:\\\/\\\/ushopwell.com\\\/ublog\\\/llms-respond-differently-to-harmful-prompts-when-ai-watermarking-is-used\\\/\"},\"wordCount\":310,\"publisher\":{\"@id\":\"https:\\\/\\\/ushopwell.com\\\/ublog\\\/#organization\"},\"articleSection\":[\"Technology\"],\"inLanguage\":\"en-US\"},{\"@type\":\"WebPage\",\"@id\":\"https:\\\/\\\/ushopwell.com\\\/ublog\\\/llms-respond-differently-to-harmful-prompts-when-ai-watermarking-is-used\\\/\",\"url\":\"https:\\\/\\\/ushopwell.com\\\/ublog\\\/llms-respond-differently-to-harmful-prompts-when-ai-watermarking-is-used\\\/\",\"name\":\"LLMs respond differently to harmful prompts when AI watermarking is used - UshopWell.com\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/ushopwell.com\\\/ublog\\\/#website\"},\"datePublished\":\"2026-09-19T02:00:27+00:00\",\"breadcrumb\":{\"@id\":\"https:\\\/\\\/ushopwell.com\\\/ublog\\\/llms-respond-differently-to-harmful-prompts-when-ai-watermarking-is-used\\\/#breadcrumb\"},\"inLanguage\":\"en-US\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\\\/\\\/ushopwell.com\\\/ublog\\\/llms-respond-differently-to-harmful-prompts-when-ai-watermarking-is-used\\\/\"]}]},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\\\/\\\/ushopwell.com\\\/ublog\\\/llms-respond-differently-to-harmful-prompts-when-ai-watermarking-is-used\\\/#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"Home\",\"item\":\"https:\\\/\\\/ushopwell.com\\\/ublog\\\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"LLMs respond differently to harmful prompts when AI watermarking is used\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\\\/\\\/ushopwell.com\\\/ublog\\\/#website\",\"url\":\"https:\\\/\\\/ushopwell.com\\\/ublog\\\/\",\"name\":\"UshopWell.com\",\"description\":\"The Premiere Online Marketplace\",\"publisher\":{\"@id\":\"https:\\\/\\\/ushopwell.com\\\/ublog\\\/#organization\"},\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\\\/\\\/ushopwell.com\\\/ublog\\\/?s={search_term_string}\"},\"query-input\":{\"@type\":\"PropertyValueSpecification\",\"valueRequired\":true,\"valueName\":\"search_term_string\"}}],\"inLanguage\":\"en-US\"},{\"@type\":\"Organization\",\"@id\":\"https:\\\/\\\/ushopwell.com\\\/ublog\\\/#organization\",\"name\":\"UshopWell\",\"url\":\"https:\\\/\\\/ushopwell.com\\\/ublog\\\/\",\"logo\":{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"https:\\\/\\\/ushopwell.com\\\/ublog\\\/#\\\/schema\\\/logo\\\/image\\\/\",\"url\":\"https:\\\/\\\/ushopwell.com\\\/ublog\\\/wp-content\\\/uploads\\\/2018\\\/01\\\/pandaSwea.png\",\"contentUrl\":\"https:\\\/\\\/ushopwell.com\\\/ublog\\\/wp-content\\\/uploads\\\/2018\\\/01\\\/pandaSwea.png\",\"width\":365,\"height\":359,\"caption\":\"UshopWell\"},\"image\":{\"@id\":\"https:\\\/\\\/ushopwell.com\\\/ublog\\\/#\\\/schema\\\/logo\\\/image\\\/\"}},{\"@type\":\"Person\",\"@id\":\"https:\\\/\\\/ushopwell.com\\\/ublog\\\/#\\\/schema\\\/person\\\/6fd1f9e0ff932e534c86c70d5acff0fc\",\"name\":\"UShopWell\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"https:\\\/\\\/secure.gravatar.com\\\/avatar\\\/4adb372cadd43b4d4c57964dab95b0f69618bf960d131c4acf49d96d6bbc9c6e?s=96&d=mm&r=g\",\"url\":\"https:\\\/\\\/secure.gravatar.com\\\/avatar\\\/4adb372cadd43b4d4c57964dab95b0f69618bf960d131c4acf49d96d6bbc9c6e?s=96&d=mm&r=g\",\"contentUrl\":\"https:\\\/\\\/secure.gravatar.com\\\/avatar\\\/4adb372cadd43b4d4c57964dab95b0f69618bf960d131c4acf49d96d6bbc9c6e?s=96&d=mm&r=g\",\"caption\":\"UShopWell\"},\"url\":\"https:\\\/\\\/ushopwell.com\\\/ublog\\\/author\\\/kburnettu\\\/\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"LLMs respond differently to harmful prompts when AI watermarking is used - UshopWell.com","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:\/\/ushopwell.com\/ublog\/llms-respond-differently-to-harmful-prompts-when-ai-watermarking-is-used\/","og_locale":"en_US","og_type":"article","og_title":"LLMs respond differently to harmful prompts when AI watermarking is used - UshopWell.com","og_description":"In response to a new European Union law, AI platforms are implementing new schemes for watermarking the content they generate. Anthropic recently disclosed its future Claude models will use SynthID-Text, an approach Google created and released as open source. It uses a secret key that subtly changes the process a model uses for choosing the next word in a sentence. Whereas a top next word choice might be \u201ccloudy,\u201d the...","og_url":"https:\/\/ushopwell.com\/ublog\/llms-respond-differently-to-harmful-prompts-when-ai-watermarking-is-used\/","og_site_name":"UshopWell.com","article_published_time":"2026-09-19T02:00:27+00:00","author":"UShopWell","twitter_card":"summary_large_image","twitter_misc":{"Written by":"UShopWell","Est. reading time":"1 minute"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"Article","@id":"https:\/\/ushopwell.com\/ublog\/llms-respond-differently-to-harmful-prompts-when-ai-watermarking-is-used\/#article","isPartOf":{"@id":"https:\/\/ushopwell.com\/ublog\/llms-respond-differently-to-harmful-prompts-when-ai-watermarking-is-used\/"},"author":{"name":"UShopWell","@id":"https:\/\/ushopwell.com\/ublog\/#\/schema\/person\/6fd1f9e0ff932e534c86c70d5acff0fc"},"headline":"LLMs respond differently to harmful prompts when AI watermarking is used","datePublished":"2026-09-19T02:00:27+00:00","mainEntityOfPage":{"@id":"https:\/\/ushopwell.com\/ublog\/llms-respond-differently-to-harmful-prompts-when-ai-watermarking-is-used\/"},"wordCount":310,"publisher":{"@id":"https:\/\/ushopwell.com\/ublog\/#organization"},"articleSection":["Technology"],"inLanguage":"en-US"},{"@type":"WebPage","@id":"https:\/\/ushopwell.com\/ublog\/llms-respond-differently-to-harmful-prompts-when-ai-watermarking-is-used\/","url":"https:\/\/ushopwell.com\/ublog\/llms-respond-differently-to-harmful-prompts-when-ai-watermarking-is-used\/","name":"LLMs respond differently to harmful prompts when AI watermarking is used - UshopWell.com","isPartOf":{"@id":"https:\/\/ushopwell.com\/ublog\/#website"},"datePublished":"2026-09-19T02:00:27+00:00","breadcrumb":{"@id":"https:\/\/ushopwell.com\/ublog\/llms-respond-differently-to-harmful-prompts-when-ai-watermarking-is-used\/#breadcrumb"},"inLanguage":"en-US","potentialAction":[{"@type":"ReadAction","target":["https:\/\/ushopwell.com\/ublog\/llms-respond-differently-to-harmful-prompts-when-ai-watermarking-is-used\/"]}]},{"@type":"BreadcrumbList","@id":"https:\/\/ushopwell.com\/ublog\/llms-respond-differently-to-harmful-prompts-when-ai-watermarking-is-used\/#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https:\/\/ushopwell.com\/ublog\/"},{"@type":"ListItem","position":2,"name":"LLMs respond differently to harmful prompts when AI watermarking is used"}]},{"@type":"WebSite","@id":"https:\/\/ushopwell.com\/ublog\/#website","url":"https:\/\/ushopwell.com\/ublog\/","name":"UshopWell.com","description":"The Premiere Online Marketplace","publisher":{"@id":"https:\/\/ushopwell.com\/ublog\/#organization"},"potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/ushopwell.com\/ublog\/?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"en-US"},{"@type":"Organization","@id":"https:\/\/ushopwell.com\/ublog\/#organization","name":"UshopWell","url":"https:\/\/ushopwell.com\/ublog\/","logo":{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/ushopwell.com\/ublog\/#\/schema\/logo\/image\/","url":"https:\/\/ushopwell.com\/ublog\/wp-content\/uploads\/2018\/01\/pandaSwea.png","contentUrl":"https:\/\/ushopwell.com\/ublog\/wp-content\/uploads\/2018\/01\/pandaSwea.png","width":365,"height":359,"caption":"UshopWell"},"image":{"@id":"https:\/\/ushopwell.com\/ublog\/#\/schema\/logo\/image\/"}},{"@type":"Person","@id":"https:\/\/ushopwell.com\/ublog\/#\/schema\/person\/6fd1f9e0ff932e534c86c70d5acff0fc","name":"UShopWell","image":{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/secure.gravatar.com\/avatar\/4adb372cadd43b4d4c57964dab95b0f69618bf960d131c4acf49d96d6bbc9c6e?s=96&d=mm&r=g","url":"https:\/\/secure.gravatar.com\/avatar\/4adb372cadd43b4d4c57964dab95b0f69618bf960d131c4acf49d96d6bbc9c6e?s=96&d=mm&r=g","contentUrl":"https:\/\/secure.gravatar.com\/avatar\/4adb372cadd43b4d4c57964dab95b0f69618bf960d131c4acf49d96d6bbc9c6e?s=96&d=mm&r=g","caption":"UShopWell"},"url":"https:\/\/ushopwell.com\/ublog\/author\/kburnettu\/"}]}},"_links":{"self":[{"href":"https:\/\/ushopwell.com\/ublog\/wp-json\/wp\/v2\/posts\/277071","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/ushopwell.com\/ublog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/ushopwell.com\/ublog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/ushopwell.com\/ublog\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/ushopwell.com\/ublog\/wp-json\/wp\/v2\/comments?post=277071"}],"version-history":[{"count":0,"href":"https:\/\/ushopwell.com\/ublog\/wp-json\/wp\/v2\/posts\/277071\/revisions"}],"wp:attachment":[{"href":"https:\/\/ushopwell.com\/ublog\/wp-json\/wp\/v2\/media?parent=277071"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/ushopwell.com\/ublog\/wp-json\/wp\/v2\/categories?post=277071"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/ushopwell.com\/ublog\/wp-json\/wp\/v2\/tags?post=277071"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}