Hook

Your team completed phishing training. They passed the quizzes. They report suspicious emails. But last week, someone installed a browser extension that summarizes customer call transcripts and sends them to a third-party API you never approved.

No credential was stolen. No malware was deployed. A helpful productivity tool just became your newest exposure vector.

Security awareness programs now compete with tools that promise speed, convenience, and intelligence. The question is no longer whether your users will click a bad link. It's whether they'll trust an AI agent they never verified.

Why This Matters Now

Traditional awareness training assumes humans are the only actors making decisions. That model breaks when employees adopt AI tools that operate autonomously, summarize sensitive data, draft code, or simulate executive voices.

Shadow AI adoption follows the same pattern as shadow IT, but with faster onboarding and less friction. A Slack bot gets installed. A ChatGPT plugin automates meeting notes. A voice assistant generates email drafts. None of these appear in your asset inventory.

By the time security teams discover these tools, data has already moved. Prompts have been logged by external vendors. Internal procedures have been exposed. The risk is not hypothetical. It is active and unmonitored.

Security leaders now need two things simultaneously: continued human-risk training and real-time visibility into which AI agents are operating inside the organization. One without the other leaves a structural gap.

Three Strategic Gaps Exposed

Productivity Tools That Operate Outside Policy

Employees adopt AI extensions, plugins, and chatbots without routing requests through IT or security. These tools integrate directly into browsers, email clients, and collaboration platforms.

• Security teams have no record of which agents are running or what permissions they hold.
• Users assume tools offered through official app stores are safe by default.
• Data flows to third-party services without classification, logging, or contractual oversight.
• Removal becomes reactive, not preventive, and often happens only after an incident.

Deepfake Attacks That Bypass Trust Signals

Voice cloning and video synthesis tools now generate realistic impersonations in minutes. Attackers use these to simulate executive requests, vendor confirmations, or urgent IT instructions.

• Training programs teach users to verify sender email addresses, but voice and video feel inherently trustworthy.
• Calls and video meetings are harder to validate than text-based phishing attempts.
• Employees lack protocols for confirming identity when the medium itself appears authentic.
• Organizations that rely solely on phishing simulations miss an entire attack category.

AI Models That Accept Sensitive Prompts Without Review

Public AI models do not filter prompts for confidentiality. Users paste customer records, internal procedures, or financial data without realizing the information may be retained or used for training.

• Organizations lose control over proprietary data the moment it enters an external model.
• Compliance teams cannot audit what was shared or with whom.
• Employees may not understand the difference between internal and public AI endpoints.
• Security policies written for file uploads do not address conversational data leakage.

The Strategic Shift Required

Security awareness training must now address two separate risk surfaces: human decision-making and autonomous agent behavior. Training alone cannot solve the visibility problem. Organizations need continuous monitoring of which AI tools are active, what permissions they hold, and what data they access.

This requires pairing education with infrastructure. Users need to understand the risks of unapproved tools. Security teams need automated detection of those tools in production. The goal is not to block all AI adoption but to ensure it happens within defined boundaries.

The shift is operational, not philosophical. Security programs that treat AI agents as invisible will continue to discover exposures after the fact. Those that instrument detection and response will surface risks before they compound.

• Extend human-risk training to include AI tool adoption and deepfake recognition.
• Deploy automated discovery for AI agents, browser extensions, and chatbot integrations.
• Establish protocols for verifying identity in voice and video communications.
• Monitor prompts and data flows to external AI services in real time.

How KnowBe4 Platform Addresses This

KnowBe4's August 2026 release adds Agent Risk Manager alongside updated training modules for phishing, password hygiene, third-party risk, and privileged access. The platform now combines user education with continuous agent monitoring.

• Productivity Tools That Operate Outside Policy: Agent Risk Manager scans environments to surface AI tools, browser extensions, and chatbots that were not formally provisioned. It provides visibility into which agents are active, what permissions they request, and where data is being transmitted. Security teams gain an inventory of autonomous tools without requiring manual reporting from users.
• Deepfake Attacks That Bypass Trust Signals: Training modules now include vishing scenarios that simulate voice and video impersonation. Phishing simulations extend beyond email to cover Microsoft Teams voicemail traps and fabricated video messages. Users learn to apply secondary verification steps when the medium feels authentic but the request is unusual.
• AI Models That Accept Sensitive Prompts Without Review: Agent Risk Manager monitors for behaviors consistent with prompt injection, data exfiltration, and unauthorized model access. It alerts on patterns where sensitive information is passed to external APIs or public AI endpoints. Detection happens in real time rather than during post-incident review.

Who This Is For

• Security awareness managers who need to cover AI adoption risks in training programs.
• Security operations teams responsible for detecting unauthorized tools and agent activity.
• Compliance managers who must audit what data leaves the organization and through which channels.
• IAM leads who need visibility into which agents hold permissions and access credentials.

Call to Action

If your security awareness program does not yet address AI agents or shadow AI, start by mapping what is already running in your environment. Visit https://knowbe4.optrics.com/platform.html

FAQ

What is Agent Risk Manager?
It is a KnowBe4 feature that discovers AI tools, browser extensions, and chatbots operating inside an organization. It provides real-time visibility into agent behavior, permissions, and data flows.

How does this differ from traditional security awareness training?
Traditional training focuses on human decision-making. Agent Risk Manager extends that model to autonomous tools that act on behalf of users. It combines education with continuous monitoring.

Does this require changes to existing workflows?
No. Agent Risk Manager operates as a discovery and alerting layer. It surfaces risks without requiring users to change how they work. Security teams decide how to respond based on detected patterns.

Can this prevent all AI-related risks?
No technology prevents all risks. Agent Risk Manager reduces exposure by surfacing what is active and where data is moving. It enables faster response when unauthorized tools or suspicious behavior are detected.