由Nvidia主导的开放安全AI联盟（OSAA）在成立仅一周后便迅速扩展至120多家成员公司[1]，并在Las Vegas举办的Black Hat大会上推出了首批AI安全建议方案[1]。该联盟成立了名为Shared AI Findings Exchange（SAFE）的工作组[1]，在AI网络安全事件的机密报告、受影响方告知和无责任分析等领域提出了建议[1]。
参与该联盟的成员包括Adobe、BlackRock、Cisco、Intel、Microsoft和Visa等企业[1]。值得注意的是，虽然OpenAI和Google曾签署了启动该组织的公开信（该信获得200多家科技公司署名），但两家公司目前并未出现在成员名单中，Anthropic同样缺席[1]。
联盟成员正积极贡献开源技术以保护AI系统[1]。其中，Nvidia提供了开源LLM漏洞扫描工具Garak[1]，Amazon贡献了agent构建工具Strands Agents和授权语言Cedar[1]。此外，Red Hat正从事agent治理工作，Okta在开发agent身份技术[1]。
The Open Secure AI Alliance (OSAA), spearheaded by Nvidia, has rapidly expanded to over 120 member companies just one week after its establishment [1]. The consortium has already unveiled its first set of security recommendations at the Black Hat conference in Las Vegas, focusing on confidential reporting of AI cybersecurity incidents, affected party notification, and no-fault analysis [1].
The alliance has established a working group called Shared AI Findings Exchange (SAFE) to coordinate these efforts [1]. Member organizations include Adobe, BlackRock, Cisco, Intel, Microsoft, and Visa, among others [1]. Notably, Anthropic and OpenAI—despite having signed the public letter that launched the initiative alongside Google and over 200 other technology companies—are currently absent from the membership [1].
Member companies are actively contributing open-source technologies to strengthen AI system protection. Nvidia has provided Garak, an open-source tool for scanning large language model vulnerabilities [1]. Amazon has contributed Strands Agents, a framework for building agents, as well as Cedar, a policy language [1]. Additionally, Red Hat is developing governance solutions for AI agents, while Okta is working on identity technologies for agent authentication [1].