Niche landing page
The missing permission layer for agents that keep working after the first yes.
Personal AI agents are moving from chat responses into browser actions, outbound messages, private-data retrieval, website publication, and recurring follow-up. A stale consent guard sits between the planner and the tool dispatcher. It asks whether the approval is still valid in the current state, and it blocks or re-prompts when the answer is no.
What the software does
It stores the original consent receipt, the evidence the user saw, the exact action scope, and the conditions that would make the permission stale. Before execution, the guard compares the live context against the receipt. If the recipient changed, the purchase amount moved, the website diff grew, the browser page navigated, or the task is outside the approved channel, the agent must ask again.
For a text message AI assistant, that means a user can approve a reply or follow-up without accidentally authorizing a different message after the thread evolves.
Evidence drift
The state the user approved no longer matches the current browser, file, account, or thread.
Scope drift
The agent is about to use a new channel, amount, recipient, domain, or capability.
Time drift
The receipt expired, the task resumed after delay, or a recurring run crossed the consent window.
Browser agents
A computer-use cache can preserve screenshots and DOM state so the guard can compare what the user approved with what the browser now shows.
Website agents
For AI agent website-building, stale consent checks should inspect source changes, generated page count, backlinks, publish branch, and live URL before deployment.