研究者复测推理提取攻击：两个月后仍可从 Astra/Sol-6.1 经第三方 API 提取推理内容
AI 导读
作者在推理提取攻击公开两个月后审计了此后引入的缓解措施，发现仍能通过第三方 API 提供商从 Astra/Sol-6.1 提取推理内容。作者已将发现披露给 OpenAI 和 Anthropic，并附上 OpenAI 关于打击协同模型蒸馏活动的公告链接。
正文
Two months after our reasoning extraction attack release, we audited mitigations introduced since then and found that we still can extract reasoning from Astra/Sol-6.1 via third-party API providers.
We disclosed our findings to OpenAI and Anthropic...
https://openai.com/index/disrupting-a-coordinated-model-distillation-campaign/