WEBVTT

1
00:00:00.150 --> 00:00:02.800
AI agents can act across your business.

2
00:00:02.900 --> 00:00:05.500
But who decides what they are authorized to do?

3
00:00:05.600 --> 00:00:08.200
Capability alone is not permission.

4
00:00:10.100 --> 00:00:13.400
Imagine an agent requesting to publish a company update.

5
00:00:13.500 --> 00:00:16.400
Its request approaches a boundary before execution.

6
00:00:20.100 --> 00:00:21.700
First, identity.

7
00:00:21.800 --> 00:00:24.300
Which agent is making the request?

8
00:00:25.100 --> 00:00:26.700
Next, scope.

9
00:00:26.800 --> 00:00:29.500
Is this action within its assigned purpose?

10
00:00:30.100 --> 00:00:31.600
Then, policy.

11
00:00:31.700 --> 00:00:34.200
Which rules govern this request?

12
00:00:35.100 --> 00:00:36.900
Finally, evidence.

13
00:00:37.000 --> 00:00:39.400
Are the required conditions supported?

14
00:00:40.100 --> 00:00:42.500
Allow an authorized request.

15
00:00:44.100 --> 00:00:46.800
Halt when required evidence is missing.

16
00:00:48.100 --> 00:00:51.000
Deny an action outside its authority.

17
00:00:52.100 --> 00:00:54.800
Or escalate for a human decision.

18
00:00:56.100 --> 00:00:59.900
Only the allowed example proceeds across the execution
boundary.

19
00:01:00.200 --> 00:01:02.000
Authorization is one event.

20
00:01:02.200 --> 00:01:04.900
The execution result is another.

21
00:01:06.100 --> 00:01:12.200
A readable receipt connects the request, authority, policy
version, decision, and reported result.

22
00:01:12.400 --> 00:01:14.300
Structured data supports inspection.

23
00:01:14.500 --> 00:01:16.800
Unknown results remain unknown.

24
00:01:17.100 --> 00:01:21.300
Our public simulation demonstrates bounded rule evaluation
with fictional inputs.

25
00:01:21.500 --> 00:01:23.800
It does not establish production enforcement.

26
00:01:24.300 --> 00:01:25.900
MilliAIre Control Plane.

27
00:01:26.100 --> 00:01:27.800
Authority before execution.