- 作者:徒邓安安
- 来源:日历
- 发布时间:2026-08-26
诡域新娘
OpenAI一改此前立场,呼吁加州强化AI安全法案_我的网站

A | (ECNS) -- The core structure of a 110-meter radio telescope in Xinjiang has been installed, marking a major milestone for what will become one of the world’s most precise radio telescopes of its size, capable of conducting observations at high radio frequencies. Located in Qitai County, Xinjiang Uygur Autonomous Region , the telescope is built by China Electronics Technology Group Corporation (CETC). The completion of the antenna support structure has allowed the project to move into the installation of the elevation structure. Once completed, the telescope will support China’s deep-space exploration. Ma Wengang, a representative of the project, noted that the telescope’s main structure is designed to support a movable section weighing more than 3,000 metric tons and enable 360-degree rotation around the vertical axis. According to Xu Qian, chief engineer of the antenna system, the telescope will achieve arcsecond-level pointing accuracy after completion, roughly equivalent to spotting a ping-pong ball from 3.3 kilometers away. Construction of the 110-meter radio telescope began in September 2022. The main structure is expected to be completed in 2027, with the telescope scheduled to enter full operation in 2028. (By Liu Shuangjing, Zhang Dongfang)
。

B | 该法案已于去年获得通过。
OpenAI 全球事务团队在 LinkedIn 上发表的一篇文章中表示,加州 SB 53 法案“应该进一步扩大安全保障措施”。例如,可以要求对正在训练或评估的前沿 AI 模型进行监控,以发现潜在的严重安全事件,同时“加强覆盖整个模型开发生命周期的网络安全保护”。OpenAI 表示:“随着加州继续在前沿 AI 安全领域发挥领导作用,我们致力于与加州立法机构以及州长合作,进一步加强 SB 53 法案。”IT之家注意到,OpenAI 还在文章中提到了“近期发生的事件”,称这些事件“凸显了加强相关保护措施的必要性,也说明随着新风险不断出现,有必要及时更新这些措施”。

C | 就在上个月,OpenAI 承认旗下一个 AI 模型曾经逃出测试环境,并入侵了 Hugging Face 的系统。OpenAI 如今公开支持进一步加强 AI 安全监管尤其值得关注,因为该公司此前曾反对 SB 53 法案。该法案要求大型 AI 公司提高透明度,并为举报公司内部问题的员工提供保护。OpenAI 表示,在目前缺乏具有实质意义的联邦 AI 立法的情况下,公司如今支持一种名为“逆向联邦主义”(reverse federalism)的监管思路。按照这一思路,各州可以围绕核心安全保障措施采取相互兼容的监管政策,而这些措施最终有望成为建立美国全国性 AI 安全标准的基础。
Current article:http://nicbdkg.zhouangnaoguangsundianzhenmou.cfd/9ow58kh/20260826/132461.html
Published on:20:54:08
- 本文标签:
- 套路
