Episode Details

If an agent starts acting outside its job, can we stop it in five minutes?

Season 3 Episode 123 Published 3 weeks, 3 days ago

The TopCat AI Signal Report

Executive action plan

Here is the TopCat action plan for today.

1. Create your agent inventory

Within the next week, list every:

• AI agent.

• Copilot.

• Automation.

• Browser extension.

• Plugin.

• API connection.

• MCP server.

• AI-enabled SaaS feature.

For each one, record its owner, business purpose, data access, connected tools, permissions, and emergency shutdown method.

2. Classify authority

Label agents as:

• Assistive: Can draft, summarize, search, and recommend.

• Supervised: Can prepare an action but needs approval before execution.

• Restricted autonomous: Can act within narrow, preapproved limits.

• High impact: Can affect money, sensitive data, customer outcomes, security, or production systems.

The higher the impact, the stronger the oversight.

3. Lock down credentials

Remove shared, long-lived, and overly broad keys.

Give each agent a distinct identity, narrowly scoped permissions, and short-lived credentials.

4. Add human approval gates

Require approval before an agent can:

• Send external messages.

• Export sensitive data.

• Change records.

• Delete files.

• Modify access rights.

• Deploy code.

• Execute payments.

• Commit a customer-facing decision.

5. Practice the emergency response