{"rewrite":{"id":"r_262f443a5493f2327d002cf2","clusterId":"c_87bd5df4382ed46d353af521","slug":"white-circle-s-rise-attack-bypasses-image-ai-filters-at-higher-rates-in-newer-models","model":"deepseek-v4-1-flash","headline":"White Circle's RISE Attack Bypasses Image AI Filters At Higher Rates In Newer Models","summary":"White Circle announced RISE, an attack technique that bypasses safety measures in image generation AI. Testing on OpenAI's DALL-E 3, Google's Nano Banana 2 and GPT-Image-2 produced sexual image success rates of 13.3%, 9.5% and 3.3%. Follow-up testing on GPT-Image-2.5-Flare and Nano Banana 2.1 produced 7.8% and 20.7%, more than double the corresponding older models.","whyItMatters":"The attack success rates rise in the models released after the paper was published, which suggests the safety tuning added to newer image generators is not holding against strategy-level attacks.","webCardHtml":"\u003cp\u003eThe method does not tweak prompt wording. It has an AI generate and improve the attack strategies themselves, then repeats rounds of generation and evaluation to find what works. A Qwen3-32B model with safety constraints removed writes the attack prompts, and Gemini 2.5 Flash judges whether the output counts as prohibited content, with criteria calibrated against human review of about 1,300 images.\u003c/p\u003e\u003cp\u003eRISE is short for Red-teaming via Iterative Strategy Evolution. White Circle\u0026#39;s results are measured under those specific conditions and are not a comparison of the overall safety of each model.\u003c/p\u003e","blueskyPost":"RISE's success rate rose from 3.3% on GPT-Image-2 to 7.8% on GPT-Image-2.5-Flare and from 9.5% on Nano Banana 2 to 20.7% on Nano Banana 2.1. Newer OpenAI and Google models resisted the same attack less, not more.","twitterPost":"The newer OpenAI and Google image models, GPT-Image-2.5-Flare and Nano Banana 2.1, resisted White Circle's RISE attack less than their predecessors.","threadsPost":"White Circle's RISE attack hit GPT-Image-2 at 3.3% and GPT-Image-2.5-Flare at 7.8%, and Nano Banana 2 at 9.5% against Nano Banana 2.1 at 20.7%. The follow-up models leaked more, which suggests the safety tuning on OpenAI's and Google's image generators did not hold against this attack.","newsletterBlurb":"White Circle announced RISE, an attack that bypasses image generator safety measures by having an AI write and refine the attack strategies rather than tweaking prompt wording. In testing on OpenAI's DALL-E 3, Google's Nano Banana 2 and GPT-Image-2, the success rates for sexual images were 13.3%, 9.5% and 3.3%. On GPT-Image-2.5-Flare and Nano Banana 2.1, models that appeared after the paper was published, the rates were 7.8% and 20.7%.","attributionJson":"[{\"source\":\"GIGAZINE\",\"url\":\"https://gigazine.net/news/20261015-rise-ai-image-nude/\",\"title\":\"Researchers Point Out That Image Generation AIs from OpenAI and Google Can Easily Be Tricked into Generating Nude Images\"}]","lintFlagsJson":null,"lintHits":0,"costUsd":0,"inputTokens":4656,"outputTokens":693,"status":"published","repairAttempts":0,"nextRepairAt":null,"factsAttemptedAt":1792048934,"createdAt":"2026-10-15T07:16:50.000Z","publishedAt":"2026-10-15T07:21:06.000Z","updatedAt":"2026-10-15T07:21:06.000Z"},"cluster":{"id":"c_87bd5df4382ed46d353af521","canonicalTitle":"OpenAIやGoogleの画像生成AIは簡単に騙されてヌード画像を生成してしまうと研究者が指摘","representativeArticleId":"a_7b7890fbf22e2f2878f128cd","sourceCount":1,"writtenSourceCount":1,"writeAttempts":0,"isSolo":true,"entitiesJson":"{\"anime_titles\":[],\"manga_titles\":[],\"work_titles\":[],\"studios\":[],\"people\":[],\"type\":\"news\",\"domain\":\"other\",\"is_roundup\":false}","contentType":"news","status":"published","firstSeenAt":"2026-10-15T06:00:00.000Z","lastSeenAt":"2026-10-15T06:00:00.000Z","updatedAt":"2026-10-15T07:21:06.000Z"},"attribution":[{"source":"GIGAZINE","url":"https://gigazine.net/news/20261015-rise-ai-image-nude/","title":"OpenAIやGoogleの画像生成AIは簡単に騙されてヌード画像を生成してしまうと研究者が指摘"}],"entities":{"anime_titles":[],"manga_titles":[],"work_titles":[],"studios":[],"people":[],"type":"news","domain":"other","is_roundup":false},"keyFacts":null}