CrowdStrike pays nearly $500 million for two AI agent security startups

The deal: two targets, two layers of defense

CrowdStrike announced on November 3, 2026 the acquisition of two AI system security startups: Fiddler AI and Dreadnode. According to reporting by the Wall Street Journal cited by TechCrunch, the company is paying about $400 million for Fiddler AI and approximately $95 million for Dreadnode, totaling nearly $500 million across the two deals.

The two startups tackle different, complementary problems. Fiddler AI, founded in 2018, built an observability platform to monitor AI agents inside organizations: it tracks an agent's actions and decisions and turns that behavior into telemetry. The company had already raised about $99 million in previous rounds, according to PitchBook data cited in the report.

Dreadnode, founded in 2023 and having raised $14 million so far, automates security testing aimed at finding flaws in models, applications and AI agents before they go into production. It's roughly equivalent to an automated pentest, but designed for risks specific to LLMs and agents: prompt injection, data leakage via connected tools, instruction manipulation, and jailbreaking.

Where each piece fits into Falcon

Both acquisitions fit into Falcon, CrowdStrike's flagship security platform, used by companies and governments. Fiddler AI becomes Falcon Observe: according to TechCrunch, the idea is to evaluate agents before deployment, track what they do in production, and feed that information as telemetry into the rest of Falcon.

Dreadnode is integrated into Falcon Guardian, a detection and response product that CrowdStrike had already been building to give visibility into what AI agents do on companies' devices. The results of Dreadnode's automated tests now feed into Falcon Guardian as security policy recommendations.

Why now: agents inside companies, without mature controls

The backdrop, according to the report, is a period of significant turbulence in the cybersecurity market: companies' internal policies and processes are lagging behind the speed at which AI is being adopted inside them. Teams and individual employees experiment with AI tools, automate workflows, vibe-code their own applications, and spin up multiple AI agents, which complicates life for whoever is responsible for security.

This mismatch is the commercial engine behind the acquisition. CrowdStrike, which already has a broad presence across companies and governments, is using its size to capture a share of a rapidly expanding market. TechCrunch notes that it's not alone: cybersecurity startups and vendors have been raising tens of millions of dollars on the same thesis, that existing security software isn't enough to monitor and control tools, connectors, and AI usage inside organizations.

What changes for those who build agents

For developers and architects putting AI agents into production in Brazil, the most direct signal from this acquisition isn't the price paid, it's market validation: two specific tool categories, agent observability and automated model red teaming, have just become off-the-shelf products from one of the world's largest corporate security vendors. That tends to speed up the moment when testing an agent's security before publishing stops being an advanced team's practice and becomes a checklist item, the way SAST and DAST already are for traditional code.

In practice, teams that today put agents into production without any observability layer, without a routine for testing prompt injection, without structured logs of which tool the agent called and with what data, are operating in the blind spot that these acquisitions are trying to cover. Before reaching for a suite like CrowdStrike's, it's possible to reduce part of that risk with simple practices:

• Log every tool call (tool call) an agent makes, along with the parameters used
• Limit the agent's permission scope by default, instead of granting broad access
• Run basic adversarial tests for prompt injection and data extraction before each deployment, the same way automated tests already run for any other feature

In the Brazilian context, this check has an extra reason behind it: automated decisions made by AI agents that affect people (credit approval, résumé screening, customer service) fall under the radar of the LGPD, Brazil's data protection law, which already provides for the right to an explanation about automated decisions. Agent observability isn't just a security issue, it's also a record the company may need to produce later.

What's still unclear

The report doesn't detail an integration timeline, nor whether Fiddler AI's and Dreadnode's products will remain available independently, outside Falcon, for customers who already use the two platforms separately. There are also no closing figures officially confirmed by CrowdStrike in the sources: the $400 million and $95 million numbers come from the Wall Street Journal's reporting, not from a company statement with finalized figures.

It remains to be seen how CrowdStrike will price this agent security package within Falcon, and whether the cost becomes accessible to smaller teams or stays restricted to large enterprise contracts.