Argus：防护AI代理攻击的开源安全平台发布

Argus是一个自托管的开源多租户安全运营平台，专门用于防护AI代理攻击1。该平台通过收集器、检测引擎、测试工具和治理代理来分析安全证据1。平台的核心设计原则是确保AI代理只能分析数据但永远无法直接执行操作，所有写入操作仅通过ActionBroker处理，AI代理只能提出待批准的行动请求1。

在技术需求方面，Argus支持Python 3.11或3.121，并兼容PostgreSQL和SQLite数据库1。对于生产环境部署，平台要求配置PostgreSQL、REDIS_URL、HTTPS、数据加密密钥和审计签名密钥，且APP_ENV必须设置为production1。在部署方式上，Argus支持Kubernetes、AWS ECS Fargate和Azure Container Apps等多种容器编排环境1。该项目采用MIT许可证发布1。

Argus, a self-hosted, multi-tenant security operations platform designed to counter AI agent-based attacks, has been released as open source software.1 The platform implements a critical security architecture in which AI agents can only analyze security evidence but cannot directly execute any operations.1 All write operations are exclusively processed through an ActionBroker component, ensuring that agents can only propose actions for approval rather than implement them autonomously.1

The platform operates with a modular architecture comprising collectors, detection engines, testing tools, and governance agents that work together to analyze security evidence.1 Argus requires Python 3.11 or 3.12 for operation and supports both PostgreSQL and SQLite databases.1 For production deployments, the platform mandates configuration of PostgreSQL, Redis URL, HTTPS, data encryption keys, and audit signing keys when the APP_ENV variable is set to production.1 The software is distributed under the MIT license.1

The platform supports deployment across multiple cloud environments, including Kubernetes, AWS ECS Fargate, and Azure Container Apps,1 enabling organizations to integrate Argus into their existing infrastructure.