本文提出面向前沿 AI 模型的部署纠正事件响应框架与实施建议,以处理危险模型被部署或已部署模型变危险的风险。
摘要:
应对 AI 模型灾难性风险的综合方法应覆盖完整的模型生命周期。本文探讨了当部署前风险管理不足时的应急计划:即要么非常危险的模型被部署,要么已部署模型变得非常危险的情形。受网络安全等行业的事件响应实践启发,我们描述了一套部署纠正工具包,AI 开发者可用其响应模型在部署后形成或被发现的危险能力、行为或使用场景。我们还提供了一个框架,帮助 AI 开发者准备并实施该工具包。最后,我们建议前沿 AI 开发者应当:(1)保持对模型访问的控制;(2)建立或扩充专门团队,以设计和维护部署纠正流程,包括事件响应计划;(3)与下游用户明确这些部署纠正属于允许采取的行动。我们还建议前沿 AI 开发者、标准制定组织和监管机构开展合作,为在事件响应中使用部署纠正确定全行业标准化方法。需要说明的是:本工作适用于通过接口(如 API)提供的前沿 AI 模型,这些接口使 AI 开发者或其他上游方能够保持对访问的控制(如 GPT-4 或 Claude);它不适用于开源模型(如 BLOOM 或 Llama-2)的灾难性风险管理,因为我们讨论的限制在很大程度上无法对其强制执行。
英文摘要:
A comprehensive approach to addressing catastrophic risks from AI models should cover the full model lifecycle. This paper explores contingency plans for cases where pre-deployment risk management falls short: where either very dangerous models are deployed, or deployed models become very dangerous.
Informed by incident response practices from industries including cybersecurity, we describe a toolkit of deployment corrections that AI developers can use to respond to dangerous capabilities, behaviors, or use cases of AI models that develop or are detected after deployment. We also provide a framework for AI developers to prepare and implement this toolkit.
We conclude by recommending that frontier AI developers should (1) maintain control over model access, (2) establish or grow dedicated teams to design and maintain processes for deployment corrections, including incident response plans, and (3) establish these deployment corrections as allowable actions with downstream users. We also recommend frontier AI developers, standard-setting organizations, and regulators should collaborate to define a standardized industry-wide approach to the use of deployment corrections in incident response.
Caveat: This work applies to frontier AI models that are made available through interfaces (e.g., API) that provide the AI developer or another upstream party means of maintaining control over access (e.g., GPT-4 or Claude). It does not apply to management of catastrophic risk from open-source models (e.g., BLOOM or Llama-2), for which the restrictions we discuss are largely unenforceable.
Frontier AI Regulation: Managing Emerging Risks to Public Safety
Centre for the Governance of AI(人工智能治理中心);Center for a New American Security(新美国安全中心);Google DeepMind(谷歌DeepMind);Brookings Institution(布鲁金斯学会);University of Virginia(弗吉尼亚大学);OpenAI;Centre for Long-Term Resilience(长期韧性中心);Centre for the Study of Existential Risk, University of Cambridge(剑桥大学存在风险研究中心);University of Washington(华盛顿大学);Convergence Analysis;Centre for International Governance Innovation(国际治理创新中心);The Andrew W. Marshall Foundation(安德鲁·W·马歇尔基金会);Edmond & Lily Safra Center for Ethics(埃德蒙与莉莉·萨弗拉伦理中心);Harvard University(哈佛大学);University of Toronto(多伦多大学);Schwartz Reisman Institute for Technology and Society(施瓦茨·赖斯曼技术与社会研究所);Vector Institute(向量研究所);Akin Gump Strauss Hauer & Feld LLP(艾金·岗波律师事务所);Cohere For AI;Microsoft(微软);Collective Intelligence Project(集体智能项目);University of California: Los Angeles(加利福尼亚大学洛杉矶分校)
When AI Fails, What Works? A Data-Driven Taxonomy of Real-World AI Risk Mitigation Strategies
Faculty of Computer Science and Engineering(计算机科学与工程系);Ss. Cyril and Methodius University(西里尔和方法ius大学);Department of Computer Science(计算机科学系);Metropolitan College(Metropolitan学院)
Orphan risks at the frontier of artificial intelligence: What diverging safety and compliance frameworks reveal about how AI companies choose the risks they prioritize
School for the Future of Innovation in Society, Arizona State University(社会创新未来学院,亚利桑那州立大学)