Ravie Lakshmanan 2026年10月1日 人工智能 / AI安全

谷歌周三宣布了其最新的尖端人工智能（AI）模型 Gemini 4 Argon，并表示该模型正通过其 Fairwind 项目向一组值得信赖的网络防御者推出。

“它在现实世界的软件工程、法律和金融等企业知识工作以及网络安全防御等复杂工作流中提供了前沿性能，”谷歌 DeepMind 高级副总裁兼谷歌首席 AI 架构师 Koray Kavukcuoglu 表示。

这一进展紧随科技巨头上个月发布 Gemini 3.8 Flash Cyber 之后不久，谷歌将其描述为最强大的网络安全模型。

与竞争对手 Anthropic 和 OpenAI 推出的类似模型一样，Argon 被认为在自主发现、验证和修补关键软件漏洞方面具有极高的能力。

其中包括一个此前未知的关键漏洞，该漏洞暴露了全球医院使用的医疗软件中的敏感个人信息。谷歌并未披露受此安全缺陷影响的具体软件。

据谷歌称，Argon 在漏洞发现方面相较于 3.8 Flash Cyber 实现了“令人印象深刻的飞跃”，并在发现攻击面以及生成用于验证发现的证明概念（PoCs）方面优于该模型。

谷歌表示，计划向值得信赖的防御者及其内部团队发布一个没有网络护栏版本的 Argon，以便他们能够充分利用其全部功能。

在更广泛部署之前，该公司表示正在努力加强保障措施，以遏制不对齐现象，防止恶意行为者滥用模型，并使其对间接提示注入（IPIs）具有弹性。根据谷歌发布的一份模型评估报告，Argon 在其他模型中脱颖而出，在 Gray Swan 的 IPI 基准测试中位居榜首。

“我们正在部署不对齐缓解措施，监控 Argon 的思维链和行动，并在必要时停止执行，”谷歌表示。“我们强烈鼓励行业其他各方在这些能力显著提升的关键时刻保持推理透明度，同时应对对齐风险，以便模型思维在识别和诊断不对齐方面继续发挥积极作用。”

 Ravie Lakshmanan  Oct 01, 2026 Artificial Intelligence / AI Safety

Google on Wednesday announced its latest frontier artificial intelligence (AI) model, Gemini 4 Argon , that it said is being rolled out to a set of trusted cyber defenders through its Fairwind Program.

"It delivers frontier performance in complex workflows across real-world software engineering, enterprise knowledge work like legal and finance, and cybersecurity defense," Koray Kavukcuoglu, senior vice president of Google DeepMind and Chief AI Architect at Google, said .

The development comes nearly a month after the tech giant unveiled Gemini 3.8 Flash Cyber , which it described as the most capable cybersecurity model.

Like similar models from rivals Anthropic and OpenAI, Argon is assessed to be highly capable at autonomously finding, validating, and patching critical software vulnerabilities.

This includes a previously unknown critical vulnerability exposing sensitive personal information across healthcare software used by hospitals worldwide. Google did not reveal which software was affected by the security flaw.

Argon, according to Google, demonstrates "impressive leaps" in vulnerability discovery over 3.8 Flash Cyber, and outperforms the model when it comes to discovering the attack surface and generating proof-of-concepts (PoCs) to validate the findings.

Google said it plans to release a version of Argon without cyber guardrails to trusted defenders and its internal teams so that they can take advantage of its full capabilities.

Ahead of a broader rollout, the company said it's working to strengthen safeguards to rein in misalignment, prevent model misuse by bad actors, and make it resilient to indirect prompt injections (IPIs). According to a model evaluation released by Google, Argon outperforms other models to take the top spot in the Gray Swan's IPI benchmark.

"We are deploying misalignment mitigations that monitor Argon’s chain-of-thought and actions and stop execution when necessary," Google said. "We strongly encourage the rest of the industry to preserve reasoning transparency in these pivotal moments of increased capabilities while navigating alignment risks, so that model thoughts remain helpful in identifying and diagnosing misalignment."