AgentTrust, in plain English
AgentTrust checks risky AI actions before they run.
When an AI agent wants to create a key, export data, approve access, send a message, or change a setting, AgentTrust can pause the action, check the rule, and record the decision.
The simple problem
AI agents can move fast. That is useful, but risky when the action touches access, money, data, users, settings, or company decisions.
Grandma version
An AI tool is about to do something important. AgentTrust says: wait, check the rule first.
Business version
Selected AI actions can pause for policy, scope, risk, expiry, and approval checks before they execute.
Developer version
Send actor, action, target, purpose, scope, and risk context. Get back a clear control result and receipt state.
AgentTrust in one story
An AI agent wants to do something sensitive. AgentTrust makes the action pass through a review checkpoint before it runs.
Before the AI acts
The point is not to block every AI action. The point is to catch selected sensitive actions before they create damage.
AI proposes“Create an API key for this vendor.”
TASK Core checksPurpose, scope, risk level, policy, and approval requirement.
Responsible reviewThe action can be approved, denied, or paused for a human reviewer.
Receipt preview
AgentTrust records the action path so the company can review what happened later.
ActionAPI key creation
RiskHigh · vendor access
DecisionReview needed
StatePending · expires in 24 hours
Where can AgentTrust help?
Start with actions that should not happen automatically without a clear checkpoint.
AI creates an API key
AI exports customer data
AI approves user access
AI sends a high-risk message
AI changes an admin setting
AI triggers a refund
AI starts a vendor workflow
AI requests emergency access
Example receipt preview
The point is not to trust the AI blindly. The point is to record the decision path before the action continues.
Trust Action Receipt
Example: an AI agent wants to create a new API key for a vendor integration.
StatusReview needed
Rule checkedAPI key creation policy
DecisionWaiting for responsible approval
Receipt statePending · expires in 24 hours
Developer view
AgentTrust should be easy to understand before it becomes API work. The technical shape is simple: send a sensitive action, get a control result, record a receipt state.
Input
- actor
- action
- target
- purpose
- scope
- risk level
TASK Core result
- allow
- deny
- review needed
- limited
- expired
- revoked
Receipt state
- request ID
- policy result
- decision state
- expiry
- review status
- verification URL
Who is this for?
AgentTrust is for teams that want AI automation, but not blind AI action.
For business teams
Use AgentTrust when AI or automation touches users, access, data, money, settings, or sensitive approvals.
For developers and security teams
Use AgentTrust with TASK Core scopes, decision receipts, expiry, revocation, and review states.
Boundary: AgentTrust does not guarantee that an AI agent is safe. It records a limited control path around selected sensitive AI-agent or automated actions. Responsible parties still make final decisions.
Try AgentTrust before you integrate it.
Start with Console, then request scoped access if the workflow fits your use case.