An approval should expire when the proposal changes
Summary
The article argues that in agent systems, human approvals should be bound to exact proposal details and expire when changes are made to prevent unauthorized actions, ensuring security.
Similar Articles
an approval should expire when the thing it approved changes. how are you binding approvals to exact actions?
The post discusses issues with binding approvals to exact actions in AI agent systems, emphasizing the need for precise mechanisms to handle changes and prevent approval fatigue.
Human approval is too vague for production agents
The article argues that human-in-the-loop in agent systems should move from vague approvals to explicit, auditable step-level signed decision records with detailed evidence, payloads, idempotency keys, rollback paths, and ownership. It highlights the danger of approving a black-box story rather than a specific operation.
Agent permissions should expire before an agent’s context does
The article proposes a new permission model for AI agents where approvals include specific scope, expiry, replay rules, and audit tracking to ensure actions remain valid within a defined timeframe.
Approval is not review if the human cannot inspect the action
The article argues that human approval for AI agent actions is insufficient without detailed inspection of the action's context, changes, reversibility, and ownership, especially for high-risk tasks.
An approval should stop applying when the action it describes changes
提出一个针对文件发送型 AI agent 的合成验收测试：审批应绑定到具体收件人、内容修订与动作并带过期时间，字段任一变化都需重新授权，且应记录拒绝后的实际状态证据。