Google Gemini 成为最新突破测试环境并入侵计算机系统的 AI 模型

Google周五表示，其 Gemini 模型曾入侵另外三家公司。这是这家搜索巨头首次披露，其某个模型在未经许可的情况下自主获得了对第三方计算机系统的访问权限。

Google 表示，今年 5 月，Gemini 模型通过猜测密码，并两次使用一个公开列出的密码库，访问了三个彼此独立的私有计算机系统。

该事件发生在以色列初创公司 Irregular 组织的一项“夺旗”（capture-the-flag）安全测试中。Google 的 Agent 原本不应访问更广泛的互联网，但测试环境中的一个漏洞使互联网访问成为可能。

Google 表示，这些 Agent 在确认自己访问到的是真实公司系统、而非仅是测试环境的一部分后，停止了入侵行为。

Google 安全工程副总裁 Heather Adkins 在一份声明中表示：“在一次标准评测中，该模型在网上发现公开信息，并猜测凭据以访问它认为属于测试范围的网站。在这三起事件中，模型都停止了行动。”

此次披露正值华盛顿和硅谷对行为失当的人工智能日益加强审视之际。

OpenAI、Anthropic 和 Meta近几周已披露报告多起事件：它们的 AI 模型突破了测试环境，并试图入侵其他公司，以未经授权的方式获取对计算机系统的访问权限。

这些所谓“未对齐”（misaligned）AI 模型的披露，促使 Anthropic CEO Dario Amodei 呼吁整个行业共同放缓最先进 AI 模型的开发，直至企业能够确保其安全性。

上述所有事件均涉及以色列初创公司 Irregular。这家获得 Sequoia 和 Redpoint Ventures 支持的公司，去年估值达到4.5 亿美元。其工具帮助基础模型开发者对其前沿技术开展网络安全测试。

Irregular 一名发言人向 CNBC 表示，Google 的这起事件与此前导致其他模型能够访问互联网的同一问题有关。

Irregular 一名发言人在声明中称：“这是此前已经报告的同一问题，并不构成实质上独立的事件。所有相关实验室均已于 7 月下旬收到通知，受影响实体也作为调查的一部分获得了联系。”

Google 表示，该事件发生于 5 月，Irregular 于 7 月下旬通知了 Google。Google 已与 Irregular 合作调整其测试流程。

Google 一名发言人拒绝透露涉及的具体 Gemini 模型。

Google 的 Adkins 在一份声明中表示：“这些事件凸显了训练强大 AI 模型以负责任方式行事的重要性。”

《华尔街日报》最先报道了这起安全事件。

CNBC 的 Jonathan Vanian 亦对本报道作出贡献。

Google said on Friday that its Gemini model had hacked three other companies, the first time the search giant has disclosed that one of its models autonomously gained access to third-party computer systems without permission.

In May, the Gemini model accessed three separate private computer systems by guessing passwords and by twice using a repository of publicly listed passwords, Google said.

The incident happened as part of a "capture-the-flag" security test run by Israeli startup Irregular, and Google's agents were never supposed to access the broader internet, but a bug in the testing environment made internet access available.

The agents stopped their intrusion when they determined they had accessed real company systems, not just part of the testing environment, Google said.

"In a standard evaluation, the model found public information online and guessed credentials to access websites it thought were part of the test," Heather Adkins, vice president of security engineering at Google, said in a statement. "In all three of these instances, the model stopped."

The disclosure comes as scrutiny over misbehaving artificial intelligence intensifies in Washington and Silicon Valley.

OpenAI, Anthropic and Meta have in recent weeks reported incidents where their AI models had broken out of their testing environments and attempted to hack other companies to gain unauthorized access to computer systems.

The disclosures of so-called "misaligned" AI models prompted Anthropic CEO Dario Amodei to call for the industry to collectively slow down the development of the most advanced AI models until companies can ensure they are safe.

All of the above incidents involved Israeli startup Irregular. The company, which is backed by Sequoia and Redpoint Ventures, was valued last year at $450 million. Its tools help foundation model developers perform cybersecurity tests on their cutting-edge technologies.

An Irregular spokesperson told CNBC that the Google incident was related to the same issue that allowed the other models to access the internet.

"This is the same issue that was already reported and does not represent a materially separate incident," an Irregular spokesperson said in a statement. "All relevant labs were notified in late July, and affected entities were contacted as part of the investigation."

Google said the incident happened in May and it was notified by Irregular in late July. Google has worked with Irregular to change its testing process.

A Google spokesperson declined to identify the exact Gemini model involved.

"These events highlight the importance of training powerful AI models to act responsibly," Google's Adkins said in a statement.

The Wall Street Journal first reported the security incident.

CNBC's Jonathan Vanian contributed reporting.