CrowdStrike Holdings Inc. plans to purchase two cybersecurity startups that focus on protecting artificial intelligence agents, the company said, pairing the deals with new capabilities for its Falcon Guardian product. The acquisitions cover Fiddler AI Inc. and Dreadnode Inc. The Wall Street Journal put the total price at $495 million, with Fiddler accounting for roughly four-fifths of that amount.
Fiddler, based in Palo Alto, California, offers a platform that developers use to evaluate AI agents for risks before deployment. The software can surface problems including malicious code produced by AI and leaks of training data. It also tests combinations of models and configurations to find the setup that keeps an agent's risk level as low as possible.
When an agent still produces prompts that violate cybersecurity rules, Fiddler lets developers inspect the reasoning workflow behind them. The company says this makes it easier to pinpoint the processing step that failed and correct it. After an agent goes into production, Fiddler relies on a set of custom models called Centor to filter malicious prompts and block harmful output, adding under 80 milliseconds of latency to reviewed requests, according to the company. Fiddler co-founder and Chief Executive Officer Krishna Gade said enterprises need to understand what their agents do, step in when they deviate, and explain why, as AI security and observability converge.
Dreadnode approaches agent security differently. Instead of tracing prompt processing workflows, it runs simulated cyberattacks to expose weaknesses, drawing on more than 100 breach tactics. It also examines the MCP servers an agent uses to do its work and ranks the issues it finds by severity so administrators can address the most urgent ones first.
CrowdStrike intends to fold Dreadnode's technology into Falcon Guardian, which maps the AI agents present in a company's technology environment and checks whether they meet cybersecurity standards. With vulnerability data from Dreadnode, the product will produce recommendations for improving security, according to CrowdStrike.
CrowdStrike also disclosed a second Falcon Guardian feature alongside the deals. The AI Kill Switch lets administrators manually shut down an AI agent that is behaving dangerously, revoke its network access and stop it from restarting. Michael Devins, a director of product marketing at CrowdStrike, described the feature in a blog post as a way to quickly disrupt dangerous AI activity when risk emerges, calling it a first step toward broader enforcement that coordinates containment across the endpoints, identities, cloud environments, SaaS applications and browsers where agents operate.
More company and startup news from TechManNews.
