# CrowdStrike plans two acquisitions to test and monitor AI agents

> Agreements for Fiddler and Dreadnode target different stages of AI system safety; both deals still need to close.

By BIGCHANGE.AI Editorial

Published: 2026-11-03T15:09:38.147Z
Updated: 2026-11-03T15:09:38.147Z
Canonical: https://bigchange.ai/blog/crowdstrike-fiddler-dreadnode-ai-agent-security

![Conceptual illustration of an analyst using a mouse while viewing generic access rows on a monitor, with a security key on the desk.](https://bigchange.ai/api/media/file/tprm-v2.png)
AI-generated conceptual illustration by BIG CHANGE.

CrowdStrike plans to acquire Fiddler and Dreadnode to add AI agent monitoring and security testing.

## The big change

- Fiddler is intended to help customers evaluate and monitor agent behavior.
- Dreadnode is intended to test models and agents for security weaknesses before deployment.
- Both deals are conditional; planned integrations do not establish tested performance.

On November 3, CrowdStrike announced separate agreements to buy Fiddler and Dreadnode. Both are expected to close in its fourth fiscal quarter of 2027, subject to customary conditions. The company plans to integrate their technology into Falcon; neither acquisition has closed.

Fiddler’s proposed role is observability: helping customers trace agent decisions and tool calls, evaluate behavior before deployment and monitor activity afterward. CrowdStrike presents those capabilities as a way to examine whether an agent acted within its intended task.

For Dreadnode, CrowdStrike describes automated tests of AI models, applications and agents, including attempts to bypass restrictions or influence agent actions. The proposed testing capability is aimed at finding weaknesses before deployment.

OWASP’s agentic security guidance describes a risk: an agent can use a legitimate tool within its authorized privileges in an unintended or unsafe way. Restricting permissions, recording activity and involving people in consequential decisions address different parts of that problem; they do not guarantee every misuse will be detected.

A separate [OpenAI monitoring system previously covered by BIG CHANGE](https://bigchange.ai/blog/openai-agent-misalignment-monitoring-tests) reviews covered conversations asynchronously. OpenAI notes that an action may already have completed before a concern is identified, so an alert does not undo it.

## Sources

- [CrowdStrike agreement to acquire Fiddler](https://crowdstrike.gcs-web.com/news-releases/news-release-details/crowdstrike-acquire-fiddler-deliver-agent-observability-ai-trust) — Proposed acquisition and monitoring.
- [CrowdStrike agreement to acquire Dreadnode](https://crowdstrike.gcs-web.com/news-releases/news-release-details/crowdstrike-acquire-dreadnode-advance-ai-trust-and-deliver) — Proposed acquisition and testing.
- [OWASP agent security guidance](https://genai.owasp.org/download/52117/?tmstv=1765059207) — Independent guidance on tool misuse.
- [OpenAI monitoring documentation](https://developers.openai.com/api/docs/guides/safety-checks/misalignment-monitoring) — Separate asynchronous monitoring boundary.