CrowdStrike is committing close to half a billion dollars to protecting AI agents that can act independently. On November 3, the company announced its intent to acquire Fiddler AI and Dreadnode for a reported combined price of approximately $495 million. Each startup addresses a different category of risk tied to autonomous agents. The size of the outlay underscores how quickly agent security has moved from a niche concern to an area where major security vendors are staking out positions.

The announcement came during CrowdStrike's Fal.Con Europe conference in Barcelona. The company highlighted the dangers posed by AI agents that can make incorrect decisions even while operating with valid credentials. Neither acquisition has closed, and CrowdStrike has not publicly disclosed the financial terms of either deal.

Two startups, one agent-security bet

The Wall Street Journal, which reported the transaction, puts the price at roughly $400 million for Fiddler and $95 million for Dreadnode.

Fiddler enables security teams to retrace an agent's decisions and tool calls, including cases where its behavior drifts away from its assigned task. CrowdStrike plans to deliver the technology as Falcon Observe, positioning it as a record of agent activity that teams can audit after the fact. Dreadnode, meanwhile, tests models and agents for vulnerabilities before deployment, and its findings will help shape Falcon Guardian policies.

Both capabilities are slated for general availability in CrowdStrike's fourth quarter of fiscal 2027. CrowdStrike's fiscal year ends in January, which places that window in late 2026 and early 2027.

CEO George Kurtz described the Fiddler deal this way: “Fiddler is the flight recorder for every agent: the observability to explain exactly what an agent did and why.”

How the Kill Switch fits in

Testing AI agents before deployment is only half the challenge. Stopping an agent that has gone haywire is the other. CrowdStrike's AI Kill Switch, which Cryptopolitan previously reported on, addresses the second problem.

The AI Kill Switch is included in Falcon Guardian. It allows any authorized administrator to deactivate rogue agents, block them from restarting on specific hosts, and cut off their network access. Every action taken through the Kill Switch is documented. The capability currently operates at the gateway level, with endpoint functionality expected in the near future.

Kurtz said: “When you deploy agents with a kill switch, you stay in control.”

CrowdStrike intends to combine three capabilities: Dreadnode's pre-deployment security testing and vulnerability identification, Fiddler's monitoring of agent behavior, and Falcon Guardian's Kill Switch to terminate rogue agent operations. The full system is not yet functional, however. Fiddler and Dreadnode still need to be integrated into the Falcon platform before the combined offering works end to end. Until then, the concrete milestones to watch are the closings of both acquisitions, the platform integration work, and the Kill Switch's expansion from the gateway to endpoints.

Why agents break containment

AI agents present a security challenge that traditional access controls cannot fully solve. Those controls define what an agent is permitted to do, but they cannot always stop it from making a poor decision once it is running.

OWASP's guidance identifies threats including prompt injection, privilege escalation, and data theft. Still, an agent does not need to be hacked to cause harm. It can misuse an approved tool or take an unintended action while operating with valid permissions.

That makes monitoring agent behavior just as important as controlling access. Yet a BCG survey found that only 23% of organizations monitor or log agent activity. CrowdStrike aims to close this gap through Fiddler, which would help security teams trace agents' decisions and pinpoint where things went wrong.

What it means for security budgets

As companies deploy more agents, protecting them could claim a growing share of security spending. Grand View Research estimates the global cybersecurity market at $302 billion in 2026, while McKinsey projects that AI-related security could grow from 4% to 15% of cybersecurity budgets within three years.

Applying those percentages to a fixed $302 billion market yields roughly $12 billion and $45 billion respectively. That calculation is an illustration, not a market forecast. AI security spans far more than agents, and actual spending patterns will change.

CrowdStrike also faces competition. Zscaler is developing zero-trust controls for agent access, and Palo Alto Networks offers Prisma AIRS alongside agent endpoint protection. That mix of network-security and platform vendors points to agent security hardening into a distinct product category of its own.

CrowdStrike's approach stands apart by combining testing, monitoring, and intervention in a single platform. For customers, the central question is whether that approach improves security without making them overly dependent on a single vendor.