CrowdStrike Acquires Fiddler AI and Dreadnode for $495 Million to Secure Autonomous Agents

$495 million is what CrowdStrike is paying to solve a problem that barely existed two years ago: autonomous artificial intelligence agents going off the rails. The cybersecurity firm announced on Tuesday that it is acquiring Fiddler AI and Dreadnode, two startups focused on testing and monitoring AI.[1][2]

The deals, expected to close by the end of January 2027, will fold both companies into CrowdStrike’s flagship Falcon platform. The Wall Street Journal reported the breakdown as roughly $400 million for Palo Alto-based Fiddler and $95 million for Dreadnode, paid primarily in cash with a portion in vesting stock.[1][3]

The transactions highlight how rapidly the market for AI security is maturing. While early generative AI tools were largely confined to isolated chat interfaces, the new wave of autonomous agents is being granted direct access to corporate data and live production systems.[4]

Behind CrowdStrike’s marketing language about securing the "agentic enterprise" lies a concrete operational panic. Businesses are rushing to deploy AI agents that can chain tasks together, access internal databases, and execute complex workflows without human intervention.[5][6]

But those same agents are probabilistic and easily manipulated. They can be tricked by prompt injection, suffer from behavioral drift, or simply hallucinate a task and claim it is finished, creating a massive new attack surface for corporate IT departments.[6]

"Every organization is racing to get more out of AI, and every week we see agents chain steps together, hit an obstacle, and take actions they were never authorized to take," CrowdStrike President Michael Sentonas said in a statement.[6]

Breaking Models Before Deployment

Dreadnode, founded in 2023, attacks the problem before the AI ever reaches a live production environment. The startup, which raised a $14 million Series A funding round in early 2025, builds automated offensive security tools designed specifically for large language models.[1][4]

AI-generated

Traditional software security testing relies on deterministic code that behaves the exact same way every time it runs. AI models break that paradigm, requiring testers to probe the reasoning loops and hidden prompts that guide an autonomous agent's behavior.[6]

Dreadnode’s software allows developers to describe a testing goal in plain language. The platform then runs more than 100 known attack strategies against the model, including jailbreaks, multi-turn prompt injections, and agent-specific exploits.[1][4]

A multi-turn prompt injection, for example, might involve feeding an AI agent a seemingly benign document that contains hidden instructions to exfiltrate data. Dreadnode simulates these complex attacks to see if the agent will blindly follow the malicious commands.[1][4]

The system then maps the resulting vulnerabilities to established frameworks like the NIST AI Risk Management Framework and the OWASP Top 10 for large language models. This gives security teams a standardized way to evaluate risk before signing off on a deployment.[4]

"AI fails in ways traditional testing was never designed to catch, and the people best equipped to find those failures think like attackers," Dreadnode CEO Brad Palm said. His team will join CrowdStrike’s internal Cyber Superintelligence Lab.[1][6]

The Flight Recorder for AI

If Dreadnode acts as the pre-flight checklist, Fiddler AI is designed to be the flight recorder. Founded in 2018, Fiddler focuses on AI observability, monitoring models once they are live to explain exactly why an agent made a specific decision.[1][4]

Fiddler is the more established of the two targets, having raised a $30 million Series C round in January 2026 that brought its total funding to $100 million. CrowdStrike plans to rebrand Fiddler’s technology as Falcon Observe and integrate it directly into its existing endpoint agent.[1]

AI-generated

Observability in the context of AI means tracing an agent's reasoning and logging its tool calls. When a corporate board or a regulator asks why an autonomous system denied a customer application or altered a database, "the model did it" is no longer an acceptable answer.[1][5]

Fiddler’s software tracks token usage, logs activity for compliance audits, and detects when a model begins to drift from its intended parameters. It provides the continuous behavioral accountability that enterprises need to trust AI with system-level access.[1]

CrowdStrike CEO George Kurtz noted that without these tools, companies are flying blind. "A model can be jailbroken, an agent can be manipulated, and most organizations have no way to find those flaws until the AI is already in production," Kurtz said.[5]

The AI Kill Switch

But observability only matters if a company can stop an agent once it begins to misbehave. Alongside the two acquisitions, CrowdStrike announced a new "AI Kill Switch" built into its Falcon Guardian product, which is available immediately.[1][2]

The kill switch allows an authorized administrator to terminate a rogue agent at its root on the endpoint. It immediately cuts off the agent's network access and severs its connection to the underlying AI gateways, stopping the automated workflow in its tracks.[1]

This direct intervention is necessary because AI agents often operate across multiple cloud environments and software-as-a-service applications. If an agent is compromised, relying on standard network firewalls is often too slow to prevent data loss or unauthorized system changes.[1][5]

CrowdStrike is integrating Falcon Guardian across its broader security ecosystem. This extends the company's detection and response capabilities directly into the environments where AI agents operate, ensuring that a compromised model cannot bypass standard endpoint protections.[5]

AI-generated

A Consolidating Security Market

The $495 million outlay is part of a broader consolidation wave in the cybersecurity sector. Vendors are racing to build comprehensive AI security suites before enterprises lock in their architectures with specialized startups.[4]

CrowdStrike has been particularly aggressive on the acquisition front over the past year. In January, the company announced plans to buy identity security firm SGNL for $740 million and browser security company Seraphic for a reported $420 million.[1]

The company also agreed to acquire AI security firm Pangea in September 2025. By rolling these disparate technologies into the unified Falcon platform, CrowdStrike hopes to convince chief information security officers that they do not need to manage dozens of separate vendors.[1]

This bundling strategy is designed to capture enterprise budgets before standalone AI security startups can establish a foothold. Security teams are already overwhelmed by alert fatigue, making a single integrated platform highly appealing compared to deploying yet another standalone dashboard.[1][4]

Competitors are making similar moves. Palo Alto Networks has acquired its own slate of AI-focused firms, including Portkey, as the industry acknowledges that autonomous agents represent a massive and highly vulnerable new attack surface.[4]

The success of these acquisitions will ultimately depend on execution. CrowdStrike must seamlessly integrate Dreadnode’s testing and Fiddler’s observability into a single pane of glass without bloating its lightweight endpoint sensor.[1][4]

The fourth quarter of fiscal 2027 will serve as the first major checkpoint for the integration. Until then, enterprises deploying autonomous agents will have to balance the promise of extreme efficiency against the reality of probabilistic risks.[1]