CrowdStrike to acquire Fiddler AI and Dreadnode for $495 million

Expanding the AI security suite

CrowdStrike Holdings Inc. has announced plans to acquire two cybersecurity startups, Fiddler AI and Dreadnode. The company intends to use the technology from these firms to strengthen its Falcon Guardian product. This move signals a major push into securing the specific infrastructure used by artificial intelligence agents.

The Wall Street Journal reported that the total cost for these acquisitions is $495 million. Fiddler AI is expected to represent the majority of this price tag, accounting for roughly four-fifths of the total sum. By bringing these companies into the fold, CrowdStrike aims to provide a more comprehensive security layer for enterprise AI.

Falcon Guardian is the primary tool that will benefit from these additions. It is designed to map out the various AI agents within a company's technology environment. The goal is to ensure that these agents adhere to established cybersecurity standards while they perform their assigned tasks.

Testing and monitoring agents

Fiddler AI specializes in helping developers identify risks before they deploy AI agents into production. Their platform allows for testing different model and configuration combinations to find the setup that minimizes risk. This includes spotting issues like training data leaks or malicious code generated by the AI.

If an agent produces a prompt that violates security rules, Fiddler provides a way to study the reasoning workflow. This helps developers identify the specific processing step where the error occurred so they can fix it. Fiddler also uses custom models called Centor to filter out malicious prompts and block harmful outputs in real time.

The Centor models are designed to be efficient, adding less than 80 milliseconds of latency to the requests they review. This speed is important for maintaining the performance of the AI while still providing a layer of protection. Fiddler co-founder and CEO Krishna Gade noted that enterprises need to understand their agents' actions and explain why they happen.

Simulating cyberattacks

Dreadnode offers a different but complementary approach to securing AI agents. Instead of analyzing internal workflows, Dreadnode launches simulated cyberattacks to find vulnerabilities. The platform uses more than 100 different breach tactics to reveal where an AI agent might be weak.

The software also looks for vulnerabilities in MCP servers, which are the components AI agents use to carry out their work. Dreadnode ranks these discovered issues by severity to help administrators prioritize their repairs. This allows security teams to focus on the most urgent threats first rather than being overwhelmed by minor issues.

CrowdStrike plans to integrate Dreadnode's vulnerability data directly into Falcon Guardian. This integration will allow the tool to generate specific cybersecurity improvement recommendations for its users. By combining these two startups, CrowdStrike offers both proactive testing and reactive threat simulation for AI systems.

The AI Kill Switch

Alongside the acquisitions, CrowdStrike revealed a new feature called the AI Kill Switch. This tool gives administrators the power to manually shut down an AI agent if it begins to exhibit risky behavior. It can also be used to strip the agent of its network access to prevent it from restarting.

The company describes this as a decisive way to disrupt dangerous AI activity as soon as it emerges. It is intended to be the first step toward broader enforcement across various environments. This includes endpoints, cloud environments, SaaS applications, and browsers where agents typically operate.

Michael Devins, a director of product marketing at CrowdStrike, explained that the goal is to coordinate containment across these different platforms. By providing a manual override, the company gives IT teams a way to stop an autonomous agent instantly. This provides a safety net for companies that are worried about AI agents acting outside of their intended scope.