WEBVTT
00:00:18.079 --> 00:00:24.719
So, what happens when an autonomous agent makes a one million error?
00:00:25.679 --> 00:00:27.679
I will tell you what happens.
00:00:28.079 --> 00:00:34.399
Everyone looks around the room and nobody can answer the question who authorized this.
00:00:35.600 --> 00:00:47.200
Last week I talked about the agentic shift, the moment we moved from AI as a tool to AI as an agent with delegated authority.
00:00:47.359 --> 00:00:56.399
Today I want to go deeper into what happens when that authority is exercised in ways we did not intend.
00:00:57.439 --> 00:01:14.719
I want to talk about drift, about shadow agents, about the accountability paradox, and about why every traditional compliance framework in existence was not designed for what we are now dealing with.
00:01:16.319 --> 00:01:21.200
Think about a credit scoring model deployed in 2023.
00:01:21.359 --> 00:01:27.200
It was trained on data from a period of relative economic stability.
00:01:28.239 --> 00:01:38.159
By 2026, interest rates, employment patterns, and spending behavior have all shifted significantly.
00:01:38.959 --> 00:01:44.719
The model is still running, but it's running on assumptions that are no longer valid.
00:01:46.079 --> 00:01:47.680
It is drifting.
00:01:48.400 --> 00:01:54.480
And every decision it makes is slightly or not so slightly wrong.
00:01:54.719 --> 00:02:02.000
The EU AI Act is not primarily a technology regulation, it is an accountability regulation.
00:02:03.200 --> 00:02:10.800
Viridian, our AI governance platform, is built to be the evidence engine for this regulation.
00:02:11.840 --> 00:02:21.439
It is saying if you deploy a high risk AI system, you are responsible for its behavior, not its creator, you.
00:02:22.240 --> 00:02:25.919
You need to be able to prove you governed it.
00:02:27.280 --> 00:02:30.080
So I want to be very direct about this.
00:02:31.280 --> 00:02:52.080
If someone in your organization has deployed an AI agent, even a simple automation, even a copilot plugin with expanded permissions, and it makes a decision that harms a customer, your organization is responsible, not the employee who deployed it.
00:02:52.400 --> 00:02:54.800
You as a company.
00:02:56.400 --> 00:03:05.280
Under the EU AI Act, under GDPR and under common law principles of vicarious liability.
00:03:06.719 --> 00:03:16.240
This gap, the gap between static compliance documents and dynamic AI behavior, is exactly why I built Viridian.
00:03:17.919 --> 00:03:23.520
Not as another document, as active governance infrastructure.
00:03:24.400 --> 00:03:28.960
Viridian monitors your AI agent continuously.
00:03:29.599 --> 00:03:33.439
It alerts you when drift exceeds thresholds.
00:03:34.080 --> 00:03:36.800
It captures evidence automatically.
00:03:37.599 --> 00:03:43.280
So when the question comes and it will come, you can answer it.
00:03:44.560 --> 00:03:49.039
Next week, I'm going to show you exactly what that looks like in practice.
00:03:51.120 --> 00:03:59.439
The accountability paradox is not unsolvable, but it requires a different approach to governance.
00:04:00.560 --> 00:04:08.159
One that is active, not passive, continuous, not annual, infrastructure, not paperwork.
00:04:09.439 --> 00:04:11.840
Next week, episode three.
00:04:20.399 --> 00:04:37.199
If you are a leader committed to building your own intentional, equitable technology strategy, you can learn more about our frameworks at Anaiyat Group by visiting anaiat.org or you can reach out to me directly on LinkedIn.
00:04:38.079 --> 00:04:41.199
Until next time, keep leading intentionally.
00:04:41.439 --> 00:04:42.240
Thank you.
00:00:18.079 --> 00:00:24.719
So, what happens when an autonomous agent makes a one million error?
00:00:25.679 --> 00:00:27.679
I will tell you what happens.
00:00:28.079 --> 00:00:34.399
Everyone looks around the room and nobody can answer the question who authorized this.
00:00:35.600 --> 00:00:47.200
Last week I talked about the agentic shift, the moment we moved from AI as a tool to AI as an agent with delegated authority.
00:00:47.359 --> 00:00:56.399
Today I want to go deeper into what happens when that authority is exercised in ways we did not intend.
00:00:57.439 --> 00:01:14.719
I want to talk about drift, about shadow agents, about the accountability paradox, and about why every traditional compliance framework in existence was not designed for what we are now dealing with.
00:01:16.319 --> 00:01:21.200
Think about a credit scoring model deployed in 2023.
00:01:21.359 --> 00:01:27.200
It was trained on data from a period of relative economic stability.
00:01:28.239 --> 00:01:38.159
By 2026, interest rates, employment patterns, and spending behavior have all shifted significantly.
00:01:38.959 --> 00:01:44.719
The model is still running, but it's running on assumptions that are no longer valid.
00:01:46.079 --> 00:01:47.680
It is drifting.
00:01:48.400 --> 00:01:54.480
And every decision it makes is slightly or not so slightly wrong.
00:01:54.719 --> 00:02:02.000
The EU AI Act is not primarily a technology regulation, it is an accountability regulation.
00:02:03.200 --> 00:02:10.800
Viridian, our AI governance platform, is built to be the evidence engine for this regulation.
00:02:11.840 --> 00:02:21.439
It is saying if you deploy a high risk AI system, you are responsible for its behavior, not its creator, you.
00:02:22.240 --> 00:02:25.919
You need to be able to prove you governed it.
00:02:27.280 --> 00:02:30.080
So I want to be very direct about this.
00:02:31.280 --> 00:02:52.080
If someone in your organization has deployed an AI agent, even a simple automation, even a copilot plugin with expanded permissions, and it makes a decision that harms a customer, your organization is responsible, not the employee who deployed it.
00:02:52.400 --> 00:02:54.800
You as a company.
00:02:56.400 --> 00:03:05.280
Under the EU AI Act, under GDPR and under common law principles of vicarious liability.
00:03:06.719 --> 00:03:16.240
This gap, the gap between static compliance documents and dynamic AI behavior, is exactly why I built Viridian.
00:03:17.919 --> 00:03:23.520
Not as another document, as active governance infrastructure.
00:03:24.400 --> 00:03:28.960
Viridian monitors your AI agent continuously.
00:03:29.599 --> 00:03:33.439
It alerts you when drift exceeds thresholds.
00:03:34.080 --> 00:03:36.800
It captures evidence automatically.
00:03:37.599 --> 00:03:43.280
So when the question comes and it will come, you can answer it.
00:03:44.560 --> 00:03:49.039
Next week, I'm going to show you exactly what that looks like in practice.
00:03:51.120 --> 00:03:59.439
The accountability paradox is not unsolvable, but it requires a different approach to governance.
00:04:00.560 --> 00:04:08.159
One that is active, not passive, continuous, not annual, infrastructure, not paperwork.
00:04:09.439 --> 00:04:11.840
Next week, episode three.
00:04:20.399 --> 00:04:37.199
If you are a leader committed to building your own intentional, equitable technology strategy, you can learn more about our frameworks at Anaiyat Group by visiting anaiat.org or you can reach out to me directly on LinkedIn.
00:04:38.079 --> 00:04:41.199
Until next time, keep leading intentionally.
00:04:41.439 --> 00:04:42.240
Thank you.
