AI-Certain
AI acts. You decide. · Deutsch

Open complete agent access test →

GUARD LAB / 0.7.0

Humans decide.

Risk-based approvals · server-side functional test

Test enforcement active
Public sandbox with sample data. You play both operator and agent. No identity verification or connected systems. The server checks each test action; the destination is simulated. Rules here are separate from browser drafts under My rules.
PENDING APPROVALS
0
Validity: five minutes
BLOCKED
0
Prohibitions take precedence
TEST RECEIPTS
0
No real transfer
Low

Read information

Automatic within the authorized task.
Medium

Change customer data

Only with the required permission.
High

Payment & export

Explicit approval required. Prohibitions remain binding.

1 · Set test rules

Rule revision 1 · Changes invalidate pending approvals.

Fixed boundary: agents cannot change their own permissions. High-risk actions always require approval.

2 · Request an action

Fixed task T-1042 · AILENDO Test. Payload and risk level come from the server-side test catalog.

Test export: first request it with the prohibition enabled. Then explicitly remove the prohibition and submit a new request. That request will wait for your approval.

3 · Review and decide

Approvals apply once to the stored action. Recipient, data scope and rule revision cannot be substituted afterwards.

No requests yet. Submit a test scenario.

Execution receipts

Only successfully checked test executions appear here.

Decision log

UTCEventID

Test data is stored in your PHP session on the server. A necessary session cookie links it to this browser. Maximum 50 requests and 100 log entries; no permanent audit archive. Session expiry or cookie deletion starts a new test run.