三巨头联手,曝谷歌、OpenAI、Anthropic 拟共同组建 AI 安全标准自律组织Three giants join forces: Google, OpenAI, and Anthropic reportedly plan to jointly form an AI safety standard self-regulatory organization


AI free token/AI免费积分
三巨头联手,曝谷歌、OpenAI、Anthropic 拟共同组建 AI 安全标准自律组织

IT之家 9 月 24 日消息,The Information 今天(24 日)晚间援引知情人士消息称,谷歌、OpenAI 和 Anthropic 三家 AI 公司正自行推进成立 AI 安全标准机构的计划,希望在没有政府监督的情况下,于今年年底或明年年初启动该组织。

知情人士称,三家公司暂定将这个自律组织命名为 Standards Authority for Frontier AI(IT之家注:直译为前沿人工智能标准管理局,下称 SAFA)。工作组部分成员曾考虑邀请多位知名人士担任 CEO,还拟定过科学顾问候选名单。

三家 AI 巨头希望建立一个面向前沿模型开发企业的行业自律组织。SAFA 计划承担多项核心职能,涵盖兼容第三方机构在模型投入使用前开展测试,并制定 AI 开发商报告安全与安保事件的方式。

参与计划的人士透露,SAFA 还将把各实验室此前自愿作出的安全与安保承诺转化为具体实践标准,同时确定独立模型及实验室审计人员需要具备的资质。

工作组成员还在讨论,SAFA 是否应直接负责模型的安全和能力测试。当下,此类测试由美国商务部人工智能标准与创新中心承担,部分工作组成员担心该中心资源不足,难以对前沿模型开展充分测试。

知情人士称,有成员主张由 SAFA 在该中心现有测试基础上增加其他测试和评估项目,但双方是否会浮现职能重叠仍不明确,工作组仍在讨论。

继续闭门商讨 SAFA 启动思路的同时,三家公司高管也在公开呼吁建立统一的安全标准,并推动更进一步的国际协调。行业之外究竟由哪些政府机构牵头制定这些标准,当下仍不明确。

本文来源:https://www.ithome.com/1/007/016.htm
若有版权问题,请来信告知:13951431913#139.com 或在右方提交评论。

IT之家 reported on September 24 that The Information, citing sources familiar with the matter, said tonight (the 24th) that Google, OpenAI, and Anthropic are independently pushing forward plans to establish an AI safety standards organization, hoping to launch the group by the end of this year or early next year without government oversight.

Sources say the three companies have tentatively decided to name this self-regulatory organization the Standards Authority for Frontier AI (noted by IT Home as literally translated as "Frontier AI Standards Authority," hereinafter referred to as SAFA). Some members of the working group had considered inviting several prominent figures to serve as CEO and had also drafted a candidate list for scientific advisors.

The three AI giants hope to establish a self-regulatory organization for frontier model development companies. SAFA is planned to take on multiple core functions, including coordinating with third-party institutions to conduct testing before models are put into use, and formulating the way AI developers report safety and security incidents.

People involved in the planning revealed that SAFA will also convert the safety and security commitments previously made voluntarily by various labs into concrete practice standards, while determining the qualifications required for independent model and lab auditors.

Working group members are also discussing whether SAFA should directly take responsibility for model safety and capability testing. Currently, such testing is carried out by the U.S. Commerce Department's AI Standards and Innovation Center, and some working group members are concerned that the center lacks sufficient resources to conduct adequate testing of frontier models.

Sources said some members advocate that SAFA add additional testing and evaluation programs on top of the center's existing tests, but whether overlapping functions will emerge between the two remains unclear, and the working group is still discussing the matter.

While continuing closed-door discussions on how to launch SAFA, executives from the three companies are also publicly calling for unified safety standards and pushing for further international coordination. Which government agencies outside the industry will take the lead in formulating these standards remains unclear at this time.

Three giants join forces: Google, OpenAI, and Anthropic reportedly plan to jointly form an AI safety standard self-regulatory organization

暂无评论,快来发表第一条评论吧!No comments yet. Be the first!

📮 需求咨询📮 Inquiry