Agent Risk Manager, KnowBe4’s AI agent security product in the KnowBe4 Platform, integrates with the Claude Compliance API to help organizations monitor Claude activity and detect risks in real time.
Today, that integration is featured in Claude’s Compliance API milestone announcement.
Agents Are Acting, but Who’s Watching?
Ask a security leader which employees used Claude last week, what they connected it to and whether any sensitive information passed through, and many won’t be able to provide a definitive answer. Not because they don’t care or because they don’t want to know, but because there hasn’t been a place for them to look.
AI agents can read files, send emails, pull records from connected systems and take multi-step action, largely outside the controls organizations already have for human activity.
KnowBe4’s 2026 From Agentic Risk to Human Wins report puts numbers on that gap: 58% of cybersecurity leaders say AI agents are already taking real actions inside their organization’s workflows, and 52% say their organization’s AI use is unapproved or ungoverned. Most security leaders know AI agents are acting on their behalf, they just haven’t been able to govern what they are doing the same way they govern their people.
We believe that human risk and AI risk are not two separate problems. They are the same problem that has grown with agents entering the workforce, and they need to be treated that way.
Our integration with the Claude Compliance API is one way we address that problem.
What Does Agent Risk Manager’s Claude Integration Do?
This integration brings an organization’s Claude activity into Agent Risk Manager, the same console security teams use to govern Microsoft Copilot Studio, ChatGPT, Google Gemini and browser-based shadow AI. With Agent Risk Manager, that activity gets inventoried, analyzed for risk and tied to the people behind the agents.
Detection built for agent risk
Agent Risk Manager’s detection agents evaluate Claude activity for prompt injection, sensitive data and PII exposure, privilege escalation, excessive agency and unapproved access, using the full context of an interaction and not just a keyword match.
See and understand Claude activity
Every connected organization gains a proactive inventory of Claude agents and tools, and a unified AI Activity Log covering detections and policy events, all in one place instead of scattered across consoles.
Coverage for your whole Claude footprint
One compliance API key is all it takes. Agent Risk Manager automatically discovers every workspace tied to your organization's key, so there's no per-team or per-agent setup to maintain. What's visible follows how you've deployed Claude. On Claude Platform, that's activity logs. On Claude Enterprise, it can also include conversation content like chats and files.
Getting Started
Getting started building an AI governance program requires being able to answer a few questions: How are you seeing AI activity today? Do you know every agent and tool currently touching your systems, or only the ones IT signed off on? If someone connected Claude to a shared drive this morning, would someone notice by the end of the day? If something did go wrong, would you know how to find out what?
We’re here to help you answer those questions.
For organizations already using Agent Risk Manager, setting up your integration with the Claude Compliance API is a simple, one time process, not an ongoing lift for IT:
- Log into the Claude console as the Primary Owner of your organization and create a Compliance API key
- Add that key in Agent Risk Manager under Setup > Integrations, using the Anthropic Claude Compliance tile
- Once the connection tests healthy, Claude activity begins flowing into the AI Activity Log and User Activity views
Availability
Agent Risk Manager and the Claude Compliance API integration are available now for select customers in early access, with expanded access coming soon.
