多篇文章讨论了确保AI系统可审计和符合法规的挑战和要求。英国审慎监管局（PRA）要求银行提供所有模型（包括AI模型）的模型治理、开发、验证和风险缓解的证据。同样，《欧盟人工智能法案》第14条要求部署者通过记录的审查点、干预能力和防篡改记录来证明人工监督。此外，由于第三方访问风险和潜在的数据泄露，不鼓励使用基于云的AI处理新闻来源等敏感数据。AI决策的可重现性仅在系统专门为此设计时才可能实现审计，因为模型和路由会随时间漂移。证明AI使用的确切来源涉及对检索到的数据块及其文档哈希进行加密签名，以创建可验证的审计跟踪。 AI
影响 强调了建立健全的治理和可验证系统以满足监管需求并确保AI部署信任的关键需求。
排序理由 该集群包含多篇讨论AI可审计性和合规性相关的监管要求和技术挑战的文章，而非单一新事件。
在 Mastodon — fosstodon.org 阅读 →
AI 生成摘要 · Google Gemini · 来自 5 个来源。 我们如何撰写摘要 →
PRA model risk and AI: what must a UK bank evidence under SS1/23? Under the PRA's SS1/23, a UK bank must evidence five things for every model, including AI: a risk-tiered model inventory, board-level governance, documented development and use, independent validation with challeng…
EU AI Act Article 14 human oversight: what must a deployer actually show? Under Article 14 of the EU AI Act, a deployer must show three things an auditor can inspect: recorded human review points before a decision takes effect, a working ability to intervene and stop, and a tampe…
Protecting journalistic sources: why a newsroom cannot use cloud AI A newsroom cannot put confidential source material into cloud AI because the service is a third party that can be compelled by a court or breached. Once material leaves the building a subpoena to the vendor, not …
Can You Re-Run an AI Decision and Get the Same Result for an Audit? Only if the system was built for it. A public cloud AI service cannot reproduce a past decision because the model and routing drift. Pinned versions, recorded inputs, fixed seeds and a sealed build let an operato…
Signed retrieval: how do you prove which source an AI actually used? You prove which source an AI used by sealing each retrieved chunk, the hash of its source document and the exact context into a cryptographically signed audit record. An auditor can then reconstruct and verify t…