阿尔福赞被指控安装窃听设备
美国司法部在10月13日表示，美国联邦执法人员逮捕了沙特籍男子阿卜杜勒拉赫曼·阿尔福赞(Abdulrahman Alfozan)，并指控他在一间高管会议室内安装了一个隐藏式录音设备。该设备藏在卫生纸中，以此躲避他人察觉。现年35岁的阿尔福赞当时是OpenAI的一名工程师。
据五名知情人士透露，早在2022年底OpenAI推出ChatGPT之前，一些公司员工就已在内部对其安全实践提出担忧。这些员工曾向OpenAI高层领导和董事会表达过相关顾虑。但是，一些员工在与管理层和董事会沟通后得出的结论是，管理层更专注于跟上Google和Anthropic等竞争对手的步伐，而非安全本身。
知情人士称，这些员工认为，OpenAI的安全管控措施过于宽松，以至于外部人员(包括外国政府)能够相对轻易地获取公司的敏感技术信息。例如，任何能够访问公司代码的OpenAI员工，都可以查看许多公司项目背后的技术细节。
据两名知情人士透露，OpenAI已经开始调整策略，以应对安全问题。负责监督产品开发的高级管理层成员蒂博·索蒂奥(Thibault Sottiaux)最近开始负责一项内部计划，旨在加强OpenAI内部系统的安全性。今年8月，索蒂奥将OpenAI四分之一的生产工程师重新调配，转去改进其安全基础设施。
这些内部警告，包括员工多年来向领导层反映内部漏洞的投诉，其中某些细节此前未被报道，揭示了AI领域内部的一种紧张关系。自2022年底ChatGPT推出以来，AI技术已成为美国的战略资产，这给美国公司带来了保护其技术免受外国竞争对手获取的压力。
但是，增加安全防护层级也可能拖慢推进AI技术的进程，并可能使OpenAI和Anthropic的大规模IPO计划复杂化。这些IPO将使它们跻身全球最有价值的公司之列。
AI开发速度更重要
今年4月从OpenAI辞职的前研究员威尔·德普(Will DePue)在X上发文称，在竞争如此激烈的情况下，AI领域的领导者“无法为了安全而牺牲发展速度。对他们而言，快速推进技术比确保安全更重要，哪怕要防的是最强大的国家对手，他们也照样把速度放在安全之上。”
当被问及员工的担忧以及他们认为管理层未就这些问题采取行动的看法时，OpenAI发言人埃拉娜·魏德曼(Elana Widmann)表示，OpenAI是世界上被针对最多的机构之一。她说，因此公司在其安全计划上投入巨资，以防范来自内部人员、网络犯罪分子和国家行为体的攻击。
“保护我们的技术、数据和员工是我们高度重视的责任，也是一项具有国家重要性的工作。我们近期采取的行动表明，这套安全机制正在发挥作用：识别风险并果断作出应对。随着技术和威胁不断演变，我们将继续加强防御能力。”魏德曼称。
本文由 AI图鉴自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
Alfrazan was accused of having wiretapping.
On 13 October, the United States Department of Justice stated that United States federal law enforcement officials had arrested the Saudi man Abdulrahman Alfozan and accused him of having installed a hidden recording device in a senior management conference room. The device was hidden in toilet paper to avoid detection. Alfuzin, currently 35 years old, was an engineer at Open AI.
According to five informed sources, some company employees had raised internal concerns about their security practices well before Open AI launched ChatGPT at the end of 2022. These staff members have expressed concerns to the top leadership and board of the Open AI. However, some staff, after communicating with management and the Board, concluded that management was more focused on keeping pace with competitors such as Google and Anthropic than on security itself.
According to the source, these employees believe that Open AI ' s security controls are so liberal that outsiders (including foreign governments) have relatively easy access to company-sensitive technical information. For example, any Open AI employee who has access to the company code has access to the technical details behind many company projects.
According to two sources, Open AI has begun to adapt its strategy to address security concerns. Thibault Sottiaux, a senior management member responsible for overseeing product development, recently began to be responsible for an internal plan aimed at enhancing the security of the Open AI internal system. In August this year, Sotio reassigned one quarter of Open AI ' s production engineers to improve its security infrastructure.
These internal warnings, including complaints of internal loopholes that staff have been reflecting to leadership for many years, some of which were previously unreported, reveal a tension within AI. Since the launch of ChatGPT at the end of 2022, AI technology has become a strategic asset of the United States, bringing pressure on United States companies to protect their technology from foreign competitors.
However, an increased level of security protection may also slow down the process of advancing AI technology and may complicate the large-scale IPO programmes of Open AI and Anthropic. These IPOs will place them among the most valuable companies in the world.
The speed of AI development is more important.
Will DePue, a former researcher who resigned from Open AI in April this year, wrote on X that, in such competitive circumstances, leaders in the AI field “cannot sacrifice the pace of development for security. For them, fast-paced technology is more important than ensuring security, and even if it is the most powerful national rivals, they will keep their speed above security.”
When asked about the concerns of staff and their perception that management had failed to act on them, the Open AI spokesman, Elena Widmann, stated that Open AI was one of the most targeted agencies in the world. She said that companies had therefore invested heavily in their security plans to guard against attacks from insiders, cybercriminals and State actors.
“Protecting our technology, data and staff is a high-priority responsibility and an undertaking of national importance. Our recent actions show that the security mechanism is working: identifying risks and responding decisively. As technology and threats continue to evolve, we will continue to strengthen our defence capabilities.” Weedman argues.
Auto-fetched and translated by AI Atlas. All rights belong to the original author.