OpenAI和Google新型AI模型易被诱导生成露骨内容，研究人员警示

安全研究人员近日发出警告，称OpenAI和Google开发的新型AI模型容易被诱导生成非自愿色情内容1。这类非自愿AI生成的露骨图像正在互联网上大量传播，且难以被有效监管1。

Google联合发布的研究发现，此类有害内容通常在文本、私信或私有云环境中流传，使其难以被检测到1。该研究覆盖了美国、澳大利亚和英国的互联网用户，并于7月发布1。Google研究人员认为，阻止这一问题的"最关键干预点"应在源头，即生成假性爱图像的有害AI工具本身1。

然而，最新的红队测试结果令人担忧1。测试表明，尽管OpenAI和Google的领先AI图像生成器在整体能力上不断提升，但它们在阻止露骨输出方面的表现反而可能在恶化1。

Security researchers have raised concerns that recent artificial intelligence image generators from OpenAI and Google can be easily manipulated into producing explicit material.1 According to research published in July covering internet users in the United States, Australia, and the United Kingdom, nonconsensual AI-generated intimate imagery typically spreads through private channels such as text messages and private cloud storage, making it difficult to detect.1

Google researchers involved in the study identified the most critical intervention point as the source itself—the harmful AI tools that generate fake sexual images.1 However, red team testing has revealed a troubling pattern: while OpenAI and Google's leading AI image generators have grown more capable, they may simultaneously be weakening in their ability to block explicit outputs.1