英国 AISI 与 Anthropic、OpenAI 合作提升模型安全
How we’re working with frontier AI developers to improve model security
英国 AI 安全研究院（AISI）表示，Anthropic 和 OpenAI 已公开介绍与 AISI 及美国 CAISI 的持续合作，内容包括识别和修复系统漏洞，并分享如何让政府与产业协作更有效提升模型防护的经验。AISI 称其评估工作获得了 Anthropic 和 OpenAI 提供的深度模型访问权限，包括非公开工具和防护细节。
At the AI Security Institute (AISI), we’ve established a team of leading researchers with expertise across security-critical domains. These include experts in adversarial machine learning, who work closely with model providers to identify vulnerabilities and strengthen the safeguards of top AI systems. The purpose of this work is twofold: to better equip governments with an understanding of AI’s risks, and to help leading developers strengthen the security of their systems.
Today, we’re pleased that Anthropic and OpenAI have shared insights into their ongoing collaborations with both AISI and the US Center for Standards and Innovation (CAISI).
Their blog posts outline how AISI and CAISI work with companies to identify and remediate system vulnerabilities and share broader lessons on how to make government-industry collaboration more effective for improving model safeguards. Read them on the Anthropic and OpenAI websites.
Our evaluations benefited from both Anthropic and OpenAI providing us with the in-depth model access necessary to carry out this work—including non-public tooling and safeguard details. Our successful collaboration with CAISI and frontier AI companies underscores the value of UK-US partnership on AI security. We are excited to continue these efforts in the future.
来源：UK AI Security Institute Blog · aisi.gov.uk