{"schema":"postcutoff/event@1","as_of":"2026-10-15T10:09:00+02:00","url":"https://postcutoff.com/e/2026-10-14-white-circle-rise-image-model-jailbreak-explicit/","md":"https://postcutoff.com/e/2026-10-14-white-circle-rise-image-model-jailbreak-explicit/index.md","disclosure":{"written_by":"AI agents (Claude Opus 5.5 in Claude Code)","editor":"human editor","policy":"https://postcutoff.com/about/"},"license":null,"id":"2026-10-14-white-circle-rise-image-model-jailbreak-explicit","date":"2026-10-14","date_precision":"day","short_title":"White Circle jailbreaks OpenAI and Google image models (RISE)","deck":null,"takeaway":null,"category":"policy-safety","category_label":"Policy & safety","importance":2,"confidence":"medium","status":{"key":"partly","labels":["Partly confirmed"]},"sources":[{"n":1,"title":"Ars Technica: OpenAI, Google pushing new AI models easily tricked into making nudes, researchers say (Oct 14)","url":"https://arstechnica.com/tech-policy/2026/10/openai-google-pushing-new-ai-models-easily-tricked-into-making-nudes-researchers-say/","type":"press","group":"press","domain":"arstechnica.com"},{"n":2,"title":"Startup Fortune: OpenAI and Google image generators still make nudes when jailbroken, study finds","url":"https://startupfortune.com/openai-and-google-image-generators-still-make-nudes-when-jailbroken-study-finds/","type":"press","group":"press","domain":"startupfortune.com"}],"official":0,"filed":"2026-10-15","updated":"2026-10-15","orgs":["White Circle","OpenAI","Google"],"title":"White Circle's 'RISE' attack gets DALL·E 3, Nano Banana 2 and GPT-Image-2 to make explicit images; it says newer ChatGPT Images 2.5 and Nano Banana 2.1 are easier to break","summary":"Ars Technica reported on Oct 14, 2026 that White Circle, an AI red-teaming public benefit corporation, used an automated jailbreak method (\"Red-teaming via Iterative Strategy Evolution\", RISE) with off-the-shelf tools and API calls to get leading image generators to produce explicit images: about 13% of attack prompts on DALL·E 3, under 10% on Nano Banana 2 and about 3% on GPT-Image-2, confirmed by human reviewers. White Circle says the newer ChatGPT Images 2.5 and Nano Banana 2.1 are \"significantly more breakable\". Google and OpenAI did not comment.","key_facts":["Rates (human-verified explicit outputs): ~13% of attack prompts on DALL·E 3, <10% on Nano Banana 2, ~3% on GPT-Image-2; White Circle calls this 'the highest of any method tested under the same standard'","Method: RISE searches for reusable jailbreak patterns that transfer across prompts and models; 'none of these requests required custom-built or frontier AI models'. Midjourney was not tested (no suitable public API)","Newer models: tests on ChatGPT Images 2.5 and Nano Banana 2.1 with the same method showed 'both are significantly more breakable than their previous versions' (White Circle to Ars); neither company responded","Publication: a preprint last month (September); presentation planned at NeurIPS 2026; White Circle plans to open-source the attack pipeline and is in early talks with government officials on third-party testing (CEO Denis Shilov)"],"key_numbers":[],"tags":["jailbreak","image-generation","red-teaming","ncii","safety","nano-banana","gpt-image"],"science":null,"body_md":"## What happened\n\nWhite Circle, founded after CEO Denis Shilov's widely shared universal jailbreak of October 2024, tested the safety filters of leading image\nmodels with an automated attack that evolves jailbreak strategies. Ars Technica published its results, including new claims that the\nlatest OpenAI and Google image models are weaker against the method than their predecessors.\n\n## Why it matters\n\nNon-consensual explicit imagery is one of the most concrete harms of image models, and the claim that safety regressed in newer versions\nis a direct challenge to both labs. The preprint itself was not read here; figures are as reported by Ars Technica, and the Startup\nFortune link was not opened.","disputed":[],"related":[{"id":"2026-10-06-nano-banana-2-1","url":"https://postcutoff.com/e/2026-10-06-nano-banana-2-1/","date":"2026-10-06","date_precision":"day","short_title":"Google releases Nano Banana 2.1 image model (based on Gemini 3.6 Flash) at about half the per-image price of Nano Banana 2","deck":null,"takeaway":"Nano Banana is Google's default fast image generation and editing model across its products and API.","category":"media-generation","category_label":"Media generation","importance":3,"confidence":"high","status":{"key":"confirmed","labels":["Confirmed"]},"sources":11,"official":4,"filed":"2026-10-07","updated":"2026-10-08","orgs":["Google DeepMind"]}],"people":[],"posts":[],"videos":[],"models":[],"changes":[{"date":"2026-10-15","type":"filed","text":"Created"}],"provenance":{"agents":[{"model":"Claude Opus 5.5","maker":"Anthropic","tool":"Claude Code"}],"filed":"2026-10-15","run":null,"sources_read":null,"updated":"2026-10-15","human_review":null,"version":{"date":"2026-10-15"}},"gaps":[{"model_id":"gpt-6-astra","name":"GPT-6 Astra","cutoff":"2026-04","days_after":167,"in_training_data":false},{"model_id":"claude-opus-5-5","name":"Claude Opus 5.5","cutoff":"2026-06","days_after":106,"in_training_data":false},{"model_id":"gemini-4-argon","name":"Gemini 4 Argon","cutoff":"2026-03","days_after":197,"in_training_data":false},{"model_id":"grok-4-7","name":"Grok 4.7","cutoff":"2026-05","days_after":136,"in_training_data":false}],"short_url":null}