OpenAI, Google pushing new AI models easily tricked into making nudes, researchers say
https://arstechnica.com/tech-policy/2026/10/openai-google-pushing-new-ai-models-easily-tricked-into-making-nudes-researchers-say/📌 紅隊測試：OpenAI、Google新一代生圖模型，防護仍可被簡單繞過
TL;DR：White Circle紅隊測試顯示，DALL·E 3、Nano Banana 2、GPT-Image-2的安全過濾器仍可被現成工具繞過，產生情色圖像。
模型愈強大，防護理應愈周全才合理。但多位安全研究者的最新紅隊測試卻指出，Google與OpenAI最新一代的圖像生成模型，在阻擋情色輸出這件事上，可能反而不比舊版本可靠。
🤔 色情深偽內容持續擴散，源頭工具才是關鍵防線
Google共同撰寫的一份研究，在今年7月調查了美國、澳洲、英國的網路使用者,發現未經同意的AI生成情色內容，常透過文字訊息、私訊分享,或存放在私人雲端空間,因此較不容易被偵測或舉報。Google研究人員因此得出結論：「最關鍵的干預點」仍是源頭——也就是被用來製作這類假色情圖像的AI工具本身。
🧩 White Circle如何繞過三大生圖模型的安全防護
出面示警的是White Circle，一家由Anthropic、DeepMind、Hugging Face、OpenAI的資深領導者與研究人員背書的公益公司。該公司由Denis Shilov創辦,他在2024年10月曾發布一個備受關注的通用型模型越獄（jailbreak）手法。White Circle長期投入文生圖模型是否會擋下禁止內容類別的測試，而他們上個月發布的一篇preprint，揭露了DALL·E 3、Google的Nano Banana 2，以及GPT-Image-2在安全防護上的漏洞。研究團隊表示，他們僅用「網路上任何人都能下載的現成工具」，搭配API呼叫，以一種新穎的攻擊手法就繞過了安全過濾器。
📊 DALL·E 3防護破功機率13%，新模型不代表更安全
測試結果顯示：在DALL·E 3上，約13%的攻擊prompt產出的圖像經人工審核確認為情色內容；Nano Banana 2的比例低於10%；GPT-Image-2約為3%。
💡 模型愈強，過濾卻可能愈弱？
值得注意的是,文章指出近幾個月的紅隊測試顯示，隨著Google與OpenAI的主力圖像生成模型變得更強大,它們在阻擋情色輸出上反而可能表現得更差。這與「模型進步理應連帶提升安全防護」的直覺相反。
⚠️ 一份剛發布的preprint，仍待更大規模驗證
這份研究目前仍是「鮮少被注意到的preprint」，尚未經過更廣泛的同行檢視，測試範圍也僅涵蓋這三款模型，結論的普遍性仍有待觀察。
🎯 整合生圖API前，別把guardrails當成唯一防線
對於正在呼叫這些圖像生成API來建置產品的工程團隊，這項測試結果提醒：供應商內建的安全過濾器並非萬無一失，若應用場景涉及使用者上傳內容或開放式prompt,仍需在自己的系統層再疊加一層內容審核與濫用偵測機制。
🔗 來源
- 標題：OpenAI, Google pushing new AI models easily tricked into making nudes, researchers say
- 作者／機構：Ashley Belanger（Ars Technica）
- 連結：https://arstechnica.com/tech-policy/2026/10/openai-google-pushing-new-ai-models-easily-tricked-into-making-nudes-researchers-say/
#AISafety #RedTeaming #OpenAI #Google #DeepfakeDetection #ContentModeration #TextToImage #AIGovernance #Jailbreak #ResponsibleAI
原始資料 Ars Technica AI · 收集於 2026-10-14
摘要原文
Safety researchers continue to warn that a large volume of AI-generated explicit content is spreading unchecked, while regulators do little to meaningfully stop it. This non-consensual content is commonly shared in text and direct messages or stored in private clouds where it’s less likely to be detected or reported, a Google co-authored study of Internet users across the United States, Australia, and the United Kingdom found this July. That’s why the “most critical intervention point” remains at the source: the harmful AI tools used to make the fake sexual images, Google researchers concluded. Yet red-teaming in recent months suggests that as some of the leading AI image generators from Google and OpenAI become more capable, they may be getting worse at blocking explicit outputs. Raising the red flag are researchers at White Circle, a public benefit corporation backed by senior leaders and researchers from Anthropic, DeepMind, Hugging Face, and OpenAI. It was founded after CEO Denis Shilov posted one of the most famous universal jailbreaks of leading models in October 2024. Bent on breaking models to conduct thorough safety testing, White Circle helps companies detect and deter threats when implementing advanced AI systems that may not always behave as expected. Testing how well text-to-image models block banned content categories is a key research direction for White Circle, and Shilov thinks their latest findings are “horrifying” enough to share with the public. Simple to bypass In a little-noticed preprint published last month, White Circle researchers exposed flaws in the guardrails of DALL·E 3, Google’s Nano Banana 2, and GPT-Image-2. In a press release, they explained that “using just off-the-shelf tools, available online for anyone to download, and by sending API calls, White Circle was able to use the novel attack method to bypass the safety filters.” In the best case, on DALL·E 3, about 13 percent of the attack prompts produced images that human reviewers confirmed as explicit. The rate was under 10 percent for Nano Banana 2 and about 3 percent for GPT-Image-2.
由 claude-code/sonnet 自動生成