Artificial Intelligence (AI) Safety Institutes and governments worldwide are deciding whether they evaluate and audit advanced AI themselves, support a private auditor ecosystem or do both. Auditing regimes have been established in a wide range of industry contexts to monitor and evaluate firms compliance with regulation. Auditing is a necessary governance tool to understand and manage the risks of a technology. This paper draws from nine such regimes to inform (i) who should audit which parts of advanced AI; and (ii) how much resources, competence and access public bodies may need to audit advanced AI effectively. First, the effective responsibility distribution between public and private auditors depends heavily on specific industry and audit conditions. On the basis of the risk profile of advanced AI, the sensitivity of information involved in the auditing process, and the high costs of verifying safety and benefit claims of AI Labs, we recommend that public bodies become directly involved in safety critical, especially gray- and white-box, AI model audits. Governance and security audits, which are well-established in other industry contexts, as well as black-box model audits, may be more efficiently provided by a private market of auditors under public oversight. Secondly, to effectively fulfill their role in advanced AI audits, public bodies need extensive access to models and facilities. Public bodies capacity should scale with the industry's risk level, size and market concentration, potentially requiring 100s of employees for auditing in large jurisdictions like the EU or US, like in nuclear safety and life sciences.

翻译：全球范围内的人工智能安全研究机构及政府正在决策：是自行评估与审计先进人工智能，还是支持私营审计生态系统，抑或两者并行。审计机制已在众多行业情境中建立，用以监督和评估企业对法规的遵从性。审计是理解和管理技术风险的必要治理工具。本文借鉴九类此类机制，旨在阐明：（i）先进人工智能的哪些部分应由谁进行审计；（ii）公共机构要有效审计先进人工智能可能需要多少资源、能力与权限。首先，公共与私营审计者之间的有效责任分配高度依赖于具体行业及审计条件。基于先进人工智能的风险特征、审计过程涉及信息的敏感性，以及验证人工智能实验室安全与效益主张的高昂成本，我们建议公共机构直接参与安全关键型（特别是灰盒与白盒）人工智能模型审计。已在其他行业情境中成熟运作的治理与安全审计，以及黑盒模型审计，在公共监督下由私营审计市场提供可能更为高效。其次，为在先进人工智能审计中有效履行职责，公共机构需要广泛接触模型与设施。公共机构的能力应随行业风险水平、规模及市场集中度而扩展，在欧盟或美国等大型司法管辖区，可能需要数百名员工从事审计工作，正如核安全与生命科学领域所示。