CrowdStrike Buys Fiddler and Dreadnode to Police AI Agents

CrowdStrike said Tuesday it has signed definitive agreements to acquire two AI-security startups, Fiddler AI and Dreadnode, in a push to make its Falcon platform the place where enterprises discover what their AI agents did, why they did it, and how to stop them when they go wrong.

The company announced both deals at its Fal.Con Europe conference in Barcelona. CrowdStrike didn't disclose a price for either acquisition. Each announcement says the purchase price will be paid predominantly in cash, with a portion delivered in stock subject to vesting conditions, and each deal is expected to close during CrowdStrike's fourth quarter of fiscal 2027, subject to customary closing conditions.

The acquisitions land at a moment when companies are deploying autonomous agents faster than they can govern them. An agent that chains steps together can take actions nobody approved, and regulators are starting to ask who was responsible. CrowdStrike is betting that observability and offensive testing for AI will become standard enterprise security spend — and that it can own the category before rivals do.

What does Fiddler bring to Falcon?

Fiddler tracks and explains what AI agents do. Once the deal closes, CrowdStrike will sell the technology as part of Falcon under the name Falcon Observe, with general availability planned for the fourth quarter of fiscal 2027.

According to CrowdStrike, Falcon Observe will:

• Evaluate agents for accuracy and policy compliance before deployment
• Trace agent decisions and tool calls across the full agent hierarchy
• Flag behavioral drift, hallucinations and unintended actions in live agents
• Track token usage and spending

CrowdStrike claims Fiddler's purpose-built models can cut evaluation costs by up to 98% without calling external LLMs. Supported frameworks include LangGraph, Amazon Bedrock, AWS Strands and Google ADK, all instrumented through OpenTelemetry.

"When a regulator or your board asks what an agent did, saying 'the model did it' is not an answer," George Kurtz, CrowdStrike's CEO and founder, said in the announcement. "Fiddler is the flight recorder for every agent: the observability to explain exactly what an agent did and why."

Krishna Gade, Fiddler's CEO and co-founder, said the startup was founded to answer one question: whether enterprises can trust what their AI agents are doing. "As AI security and observability converge, enterprises need to understand their agents' actions, intervene when they go off course, and explain why," he said.

The framing matters beyond marketing. Audit trails for automated decisions are becoming a compliance requirement in regulated industries, and observability vendors that can't prove what an agent did at the level of individual tool calls will struggle to pass board-level scrutiny.

What is Dreadnode for?

Dreadnode covers the earlier stage of the lifecycle: attacking AI models and agents to find weak spots before they reach production. Its software runs more than 100 known AI attack strategies — among them jailbreaks, prompt injection and multi-turn, agent-specific attacks — against models, agent reasoning loops, applications and the MCP servers that connect agents to tools.

Findings are ranked by severity and mapped to established frameworks:

• OWASP Top 10 for LLMs
• MITRE ATLAS
• NIST AI RMF

CrowdStrike said those findings will feed Falcon Guardian, its AI detection and response product, as policy recommendations. Dreadnode's team will join the company's Cyber Superintelligence Lab. The testing product is also slated for the fourth quarter of fiscal 2027.

"AI fails in ways traditional testing was never designed to catch, and the people best equipped to find those failures think like attackers," said Brad Palm, Dreadnode's CEO.

The logic is straightforward red-teaming applied to agents. If a prompt injection can make an agent call the wrong tool, CrowdStrike wants that failure found in a lab, not in production.

What is the AI Kill Switch?

One piece of Tuesday's news didn't have to wait for a deal to close. CrowdStrike released an AI Kill Switch in Falcon Guardian that lets an authorized administrator stop a misbehaving agent immediately.

The kill switch can:

• Stop a misbehaving agent
• Block it from relaunching on a host or host group
• Cut its enterprise network access
• Log every action taken

Termination happens first at the Gateway and "soon" on the endpoint, the company said.

"One control point isn't enough to stop that," Michael Sentonas, CrowdStrike's president, said in a separate release on the Guardian updates. He was describing agents that chain steps together and take actions they were never authorized to take.

How does this fit the wider agent-security race?

CrowdStrike isn't the only vendor building agent controls, and Tuesday's announcements position it against several competitors at once. IBM recently tied AI agent identity to HashiCorp Vault so access can be cut off mid-task. Fortinet added prompt and MCP tool-call inspection to FortiGate through a cloud service.

CrowdStrike is also reaching into Anthropic's product line. The company said Tuesday it is bringing Falcon Guardian natively into Claude Cowork, Anthropic's agent product. The Next Web reported that the integration is in beta on macOS, with Windows support expected later in November.

The pattern across the industry is consistent: as agents gain the ability to take real actions — spending money, calling tools, modifying systems — security vendors are racing to supply the identity, inspection, testing and emergency-stop layers that traditional endpoint protection never needed.

What comes next?

Fal.Con Europe wraps up Wednesday. On Monday, its first day, CrowdStrike announced security deals with Dell, Dynatrace and CoreWeave, according to The Next Web.

The conference has effectively become a rolling product launch for CrowdStrike's AI security stack: observe the agent, attack it in testing, guard it in production, and kill it when necessary. With both acquisitions expected to close in the fourth quarter of fiscal 2027 and Falcon Observe and the Dreadnode testing product slated for the same window, CrowdStrike is aiming to have a full agent-governance suite in market within a year — and to settle the question of whether "the model did it" will ever again count as an answer.