CrowdStrike plans to acquire Fiddler AI and Dreadnode and has added an AI Kill Switch to Falcon Guardian, extending its platform from stopping agents at runtime to testing them before release and explaining their behavior afterward. The package hangs together, but halting an agent’s process is not the same as revoking its authority, and the explanations will have to prove they are faithful.
What Is Covered in This Article:
- Two AI deals at Fal.Con Europe: CrowdStrike agreed to buy Fiddler AI (agent observability, to become Falcon Observe) and Dreadnode (automated AI security testing), for about $495 million combined according to The Wall Street Journal.
- A loop, not three products: Testing findings are meant to become runtime policy, and runtime behavior can then be traced and explained, across frameworks through OpenTelemetry.
- The authority gap: The new AI Kill Switch stops an agent’s process, but containing an agent also means revoking its grants and accounting for actions it already took.
- A crowded layer: Cisco, Dynatrace, and Palo Alto Networks are buying the same observability and evaluation capabilities, while security buyers rank outsourced AI red-teaming last among their AI-security priorities.
- New buyer, new seam: Falcon Observe brings ML and platform teams into Falcon, which raises the question of who owns agent behavior.
The News: At Fal.Con Europe 2026 on November 3, CrowdStrike announced definitive agreements to acquire two AI startups, Fiddler AI and Dreadnode. Terms were not disclosed beyond a mostly cash payment with some vesting stock. The Wall Street Journal reported a combined price of about $495 million, roughly $400 million for Fiddler and $95 million for Dreadnode. Both deals are expected to close in CrowdStrike’s fiscal fourth quarter of FY27, which ends in January 2027. Fiddler will be sold as Falcon Observe and Dreadnode’s technology as Automated AI Security Testing, both slated for general availability in that quarter, and the Dreadnode team will join CrowdStrike’s Cyber Superintelligence Lab.
Fiddler, founded in 2018 and based in Palo Alto, California, started in machine-learning explainability and now offers evaluation, tracing, and drift detection for AI agents, instrumented through OpenTelemetry. It has raised $100 million in total, including a $30 million Series C in January 2026. Dreadnode, founded in 2023 by former AI red-team researchers from NVIDIA and NetSPI, automates adversarial testing of models, agents, applications, and Model Context Protocol (MCP) servers, mapping findings to OWASP, MITRE ATLAS, and NIST frameworks. It raised a $14 million Series A in 2025.
CrowdStrike also added an AI Kill Switch to Falcon Guardian, generally available since September 2026. The Kill Switch terminates unauthorized agents at CrowdStrike’s AI Gateway today, with endpoint enforcement to follow, and blocks them from running again across the organization. Other Guardian additions include hooks inside coding and productivity agents and AI traffic inspection on the endpoint.
The deals extend a steady acquisition run for the Austin-based company: Humio (2021), Bionic (2023), Flow Security and Adaptive Shield (2024), Onum and Pangea (2025), and SGNL and Seraphic Security in January 2026, plus a purchase of XM Cyber’s intellectual property this year.
CrowdStrike Buys Fiddler and Dreadnode to Test, Watch, and Stop AI Agents
Analyst Take: When a board, an auditor, or a regulator asks what an agent did, and why, who in the organization can answer? CrowdStrike is betting the answer should come from the security platform. Its stated thesis for the two deals is about AI safety as much as AI security: Falcon Guardian protects agents from attack and misuse, Fiddler (as Falcon Observe) shows whether they behaved as intended, and Dreadnode tests models and agents before they ship. The company calls Fiddler “the flight recorder for every agent.”
We see a second motive in the shape of the deals, compatible with the first. Agent observability has been consolidating into larger security and observability vendors through 2026, and Fiddler was among the more established independents still standing. Both deals also bring a buyer into Falcon that CrowdStrike has not traditionally sold to: the machine-learning (ML) and platform teams that build and run AI applications, not the security operations center.
Testing, Runtime, and Explanation in One Loop
The most interesting part of the package is the plumbing between the pieces. Dreadnode findings are meant to feed Guardian as policy recommendations, so a jailbreak found before deployment becomes a runtime control after it, and Fiddler’s traces can then show whether that control held. Fiddler instruments agents through OpenTelemetry and supports frameworks such as LangGraph, Amazon Bedrock, and Google’s Agent Development Kit, which matters for enterprises running more than one agent stack.
The endpoint sensor is the other asset in play, one CrowdStrike shares with other endpoint security vendors but that network-only approaches lack. An agent running a local or open-weight model may never touch a gateway, and the sensor already sits on the device where that agent executes.
A Kill Switch Stops the Process, Not the Authority
The AI Kill Switch will draw the most attention, and it is where we would temper expectations. Terminating a rogue agent and blocking it from running again treats the agent as a process to contain. Agents, however, act through authority: delegated credentials, OAuth grants, and tokens issued to tools and MCP servers.
Stopping the process does not by itself revoke that authority, and it does nothing for actions that have already landed in a software-as-a-service (SaaS) application, a payment system, or a code repository. To us, containing an agent means containing its authority and designing for reversibility, a problem for the whole category, not one CrowdStrike created.
Explanation carries a similar caveat. A trace of tool calls and reasoning steps shows what an agent did; whether a model’s stated reasoning reflects why it acted is still an open research question, and an explanation that is plausible but unfaithful could do more harm than none in front of an auditor. The underlying need is clear enough: 44% of security decision-makers in Futurum’s 1H 2026 Cybersecurity Decision-Makers Survey (N=929) agree they struggle to baseline normal behavior for their AI models, to the point where telling a hallucination from a prompt injection attack is nearly impossible.
There is also an organizational seam. Fiddler’s existing users sit largely in data science, ML platform, and risk teams, while Falcon’s sit in security operations. Selling Falcon Observe through the platform assumes those groups will share a tool and an owner for agent behavior, which many organizations have not settled.
Security and Observability Vendors Are Buying the Same Layer
CrowdStrike is far from alone in reaching for this layer. Cisco (Galileo, folded into Splunk) and Dynatrace (Arize) bought agent evaluation and observability specialists in 2026, and Palo Alto Networks paired an observability acquisition with earlier AI-security deals. On the testing and guardrails side, SentinelOne, Check Point, and Zscaler each bought AI-security startups in 2025, while specialists such as HiddenLayer remain independent and well funded.
Buyer appetite for the testing half is less settled. In the same survey (N=904), contracting external firms for continuous AI red-teaming was the least-cited of eight options for architecting and funding AI security over the next 12 to 18 months, at 29% of top-three mentions, while relying on an existing extended detection and response (XDR) or cloud security platform with no net-new vendors drew 35%. We read that as a mild tailwind for CrowdStrike’s approach: organizations unlikely to hire outside red teams may accept automated testing that arrives inside a platform they already run.
Identity Is Where the Authority Gap Closes
Several of the pieces needed to close the authority gap already appear on CrowdStrike’s roadmap. The Guardian release lists agentic identity controls as a coming capability, and the SGNL acquisition earlier this year brought just-in-time access. Tying a kill to the revocation of an agent’s grants, and logging what the agent had already changed, would move the switch much closer to containment. Until Falcon Observe and the testing product ship, customers are buying the runtime half and a roadmap for the rest.
Questions for the Proof of Concept
In a proof of concept, ask what the AI Kill Switch revokes when it fires (the process, the agent’s tokens and grants, or both) and what record it leaves of actions the agent had already taken.
At contract time, settle who owns Falcon Observe. If the AI platform team is the daily user and the security team holds the budget, decide that before the tool arrives, ideally well before the first incident.
The hurdle for CrowdStrike is to show that Falcon Observe’s explanations hold up under scrutiny, faithful to what the agent did rather than a plausible story, since that is the evidence an auditor or a board will rely on.
What to Watch:
- When does endpoint enforcement arrive? The Kill Switch works at the AI Gateway today, with the endpoint promised “soon.” A ship date, and Mac parity with Windows for endpoint network inspection, will show how fast the runtime half fills in.
- Will Fiddler stay sold outside Falcon? Fiddler has customers that do not run CrowdStrike. Whether a standalone offering survives the close, or everything moves into Falcon Observe, will shape how those customers react.
- How will Falcon Observe be priced? A separate module, a Guardian bundle, or Falcon Flex consumption would each signal whether CrowdStrike sees agent observability as a security line item or an AI-platform one.
- Who buys the remaining independents? With Galileo, Arize, and now Fiddler gone, watch HiddenLayer, Zenity, and other AI-security specialists, and whether Microsoft or Google respond with deals of their own.
For more information, read the full announcement from CrowdStrike.
Read the Dreadnode acquisition announcement and the Falcon Guardian AI Kill Switch announcement.
Declaration of generative AI and AI-assisted technologies in the writing process: This content has been generated with the support of artificial intelligence technologies. Due to the fast pace of content creation and the continuous evolution of data and information, The Futurum Group and its analysts strive to ensure the accuracy and factual integrity of the information presented. However, the opinions and interpretations expressed in this content reflect those of the individual author/analyst. The Futurum Group makes no guarantees regarding the completeness, accuracy, or reliability of any information contained herein. Readers are encouraged to verify facts independently and consult relevant sources for further clarification.
Disclosure: Futurum is a research and advisory firm that engages or has engaged in research, analysis, and advisory services with many technology companies, including those mentioned in this article. The author does not hold any equity positions with any company mentioned in this article.
Analysis and opinions expressed herein are specific to the analyst individually and data and other information that might have been provided for validation, not those of Futurum as a whole.
Read the full Futurum Group Disclosure.
Other Insights From Futurum:
Microsoft Copilot Harness Preview Keeps Tool Approval Policy With Microsoft
Salesforce Moves to Acquire Cresta
PagerDuty’s Paige Bets Agentic AI Can End the 3 a.m. Page
Author Information
Fernando Montenegro serves as the Vice President & Practice Lead for Cybersecurity & Resilience at The Futurum Group. In this role, he leads the development and execution of the Cybersecurity research agenda, working closely with the team to drive the practice's growth. His research focuses on addressing critical topics in modern cybersecurity. These include the multifaceted role of AI in cybersecurity, strategies for managing an ever-expanding attack surface, and the evolution of cybersecurity architectures toward more platform-oriented solutions.
Before joining The Futurum Group, Fernando held senior industry analyst roles at Omdia, S&P Global, and 451 Research. His career also includes diverse roles in customer support, security, IT operations, professional services, and sales engineering. He has worked with pioneering Internet Service Providers, established security vendors, and startups across North and South America.
Fernando holds a Bachelor’s degree in Computer Science from Universidade Federal do Rio Grande do Sul in Brazil and various industry certifications. Although he is originally from Brazil, he has been based in Toronto, Canada, for many years.
