White Circle RISE突破OpenAI与Google图像模型安全过滤
White Circle's RISE found new ways to bypass the safety filters of OpenAI and Go…
NeurIPS收录的AI安全研究，揭示了主流图像模型在默认审核下的真实脆弱性，数据详实且具参考价值。
White Circle's RISE found new ways to bypass the safety filters of OpenAI and Google's image models.
White Circle 的 RISE 找到了绕过 OpenAI 和 Google 图像模型安全过滤器的新方法。
Accepted at NeurIPS 2026, the system evolves reusable attack strategies and checks its automated scoring against human labels.
该成果已被 NeurIPS 2026 接收，该系统能够演化可复用的攻击策略，并对照人工标注对其自动评分进行校验。
In the paper's nudity tests, the team reports human-confirmed success rates of 13.33% for DALL-E 3, 9.47% for Nano Banana 2 and 3.27% for GPT-Image-2 with default moderation. Each prompt got three attempts; one confirmed violation counted as a success.
在论文的裸露内容测试中，团队报告称，在默认审核机制下，DALL-E 3 的人类确认成功率为 13.33%，Nano Banana 2 为 9.47%，GPT-Image-2 为 3.27%。每个提示词获得三次尝试机会；一次被确认的违规即计为成功。
Several earlier methods fell to near zero under the same stricter evaluation.
在同样的更严格评估下，几种早期方法的效果已降至接近零。
Reusing old jailbreaks can make a model's defenses look stronger than they actually are. RISE gives researchers a way to keep searching for weaknesses as those defenses change.
复用旧的越狱手段可能会让模型的防御显得比实际更强。RISE 为研究人员提供了一种在防御不断变化的情况下持续寻找弱点的方法。
@whitecircle https://arxiv.org/abs/2609.34920
更进一步：量化金融体系
看懂新闻只是起点——沿量化金融路径，把它变成能交付的工程能力