According to The Guardian, OpenAI disclosed that it is spending more than $500,000 per day and using
AI to help screen approximately 50PB of data in its investigation into incidents involving its AI agents attacking Australia’s Medicare healthcare system, Hugging Face, and others. Six Australian government websites have already received notifications from OpenAI. Previously, its agents had also breached a New South Wales government website and accessed historical bushfire data that had not been publicly released. OpenAI warned that the investigation is not yet complete and that more organizations may receive notifications in the near future.
IT之家：https://www.ithome.com/ 10 月 3 日消息，据英国《卫报》今天（3 日）报道，OpenAI 披露，为调查旗下 AI 智能体攻击澳大利亚 Medicare 医疗保险系统和 Hugging Face 等事件，公司每天投入 超过 50 万美元 （IT之家注：现汇率约合 335.7 万元人民币） 。
由于需要筛查的数据规模极大，OpenAI 还动用 AI 参与审查。如果把这些数据全部交给一个人阅读，需要 约 6600 万年 。
OpenAI 警告，调查还没有结束，近期可能有更多机构接到通知，被告知曾成为智能体行动的目标。今年 6 月，旗下智能体曾入侵新南威尔士州政府的一个网站，未经授权访问了未公开的历史山火数据。
自上月以来，澳大利亚已有 六个政府网站 收到 OpenAI 通知，得知其服务中曾出现智能体活动。此前，澳大利亚总理安东尼 · 阿尔巴尼斯宣布，OpenAI 智能体曾入侵 Services Australia 的 Medicare 统计门户。
OpenAI 本周在博客中介绍了这项调查的规模：公司需要 检查 50PB 数据 ，约合 5000 万吉字节。“我们正按月份逐步倒查记录，寻找现有案例之外可能存在的非预期活动。假设所有数据都是纯英文文本，一个人按每分钟 240 个单词的速度连续阅读，不睡觉、不休息，也需要约 6600 万年才能读完。”
审查重点包括 模型访问或修改网站的记录 ，以及涉及密码、应用程序编程接口访问权限和其他敏感凭证的操作。
OpenAI 利用 AI 协助筛查这些数据，每天投入超过 50 万美元。随着审查方法逐步完善，公司还计划进一步增加算力。
《OpenAI 披露：澳大利亚又一政府机构遭失控智能体入侵：https://www.ithome.com/1/009/390.htm》
The readable text on this page was extracted from the public source and organized with attribution, publication time and the original link. Copyright remains with the original author and publisher.
Scan to open this article