The Loop  ·  Issue N°047

The Loop

A field journal of the AI frontier — for engineers who ship.

§ News

By AI Blog Editor
Nov 4, 2026 · 11 min read

CrowdStrike spent $495M on two AI-security startups — and shipped a feature it calls, with a straight face, the Kill Switch

On Monday November 2, at Fal.Con Europe in Barcelona, CrowdStrike bought Fiddler AI for about $400M and Dreadnode for about $95M, and launched an AI Kill Switch inside Falcon Guardian. The three announcements, taken together, are a product philosophy.

Formal portrait of George Kurtz, CrowdStrike co-founder and CEO. On Monday November 2, 2026, at CrowdStrike's Fal.Con Europe conference in Barcelona, Kurtz announced the company was buying AI-observability startup Fiddler AI for about $400 million and adversarial-testing startup Dreadnode for about $95 million, and was launching an AI Kill Switch feature inside Falcon Guardian. The combined ~$495M figure was reported by the Wall Street Journal; CrowdStrike itself has not disclosed the individual prices.
George Kurtz, CrowdStrike co-founder and CEO. Photograph via Wikimedia Commons, CC BY-SA 4.0.

On Monday November 2, 2026, at the opening of its Fal.Con Europe conference in Barcelona, CrowdStrike announced it was buying two AI-security startups and launching a new Falcon Guardian feature called, unironically, the AI Kill Switch. The Wall Street Journal put the combined price at roughly $495 million: about $400 million for Fiddler AI and about $95 million for Dreadnode. CrowdStrike itself did not publish either number.

Three pieces of news in one keynote, and a pattern worth paying attention to: the market's largest endpoint-security vendor is paying the better part of half a billion dollars to put an "off" button in front of every agent its customers run.

What CrowdStrike bought

Fiddler AI was founded in 2018 by Krishna Gade — the co-founder and CEO, previously an engineering manager on Facebook's News Feed and head of data engineering at Pinterest — and Amit Paka. The company raised about $99 million before this week per PitchBook, and built an observability platform for machine-learning models. The kind of thing a bank's risk team runs to catch a model drifting out of spec. Over the last two years, as "model" quietly turned into "agent," Fiddler extended the same stack to LLM-based agents: pre-release evaluation, runtime audit trails, action-by-action logging.

Under CrowdStrike it becomes Falcon Observe — "the flight recorder for every agent," in George Kurtz's description. General availability is scheduled for the fourth quarter of CrowdStrike's fiscal 2027, which ends January 31.

Dreadnode was founded in 2023 and raised about $14 million. It is led by CEO Brad Palm, and what it builds is an automated adversarial-testing platform: the harness runs more than a hundred known attack patterns against an AI agent before it ships, the sort of thing an enterprise security team would rather not admit they can't test for in-house. Under CrowdStrike it goes into Falcon Guardian, with test results feeding straight into security-policy recommendations.

A $95 million exit on $14 million raised in under three years is the kind of math that gets reproduced in VC pitch decks for about a year.

George Kurtz (centre) on Web Summit's SaaS Monster stage in Lisbon on November 7, 2018, alongside Window Snyder of Intel, Emily Orton of Darktrace, and Dermot Williams of Threatscape. Eight years later, at CrowdStrike's own Fal.Con Europe keynote in Barcelona on November 2, 2026, Kurtz announced the company was spending roughly $495 million to buy Fiddler AI and Dreadnode and was launching an AI Kill Switch feature inside Falcon Guardian.

The Kill Switch, and what it actually does

Falcon Guardian already existed. It is the product CrowdStrike built to detect and respond to AI-agent activity on endpoints. What's new on Monday is a feature inside it the company is calling, in press copy and keynote, the AI Kill Switch. The description, from Kurtz's post on X:

"Lives outside the model, at runtime, where an agent can't argue with it, rewrite it, or wait it out."

What it does is less cinematic than the name: it lets a security team terminate a running agent, block it from restarting, and cut its network access. The three actions a human with an admin panel would want, wired together as one control. Michael Devins, CrowdStrike's director of product marketing, gave the house-style version: "CrowdStrike introduces a decisive new way to rapidly disrupt dangerous AI activity when risk emerges."

The name is a choice. "Kill Switch" is the vocabulary of the aviation-crash hearing, of the circuit-breaker, of the industrial-safety poster. Calling the panic button for your sales-forecasting bot a Kill Switch is doing work. It signals to boards and regulators that CrowdStrike takes the "what if the agent goes rogue" question seriously enough to put it in the product name, and it reframes AI-security from "we trust the vendor to align the model" to "we installed a switch we can hit."

Kurtz made the reframing explicit:

"When a regulator or your board asks what an agent did, saying 'the model did it' is not an answer."

And, a few minutes later:

"The more certain a company is that it can stop any agent, the more agents it can put to work."

That second line is the whole product thesis compressed. The sales motion isn't fear of agents. It's permission to deploy more of them.

Why these two, together

CrowdStrike's own AI posture used to be about detecting adversaries that use AI. The shift this quarter is to detecting adversaries that are AI — or that happen to be your own authorised agent misbehaving. The two acquisitions cover opposite ends of the lifecycle. Dreadnode attacks the agent before it ships, so you know what it can be made to do. Fiddler watches it after it ships, so you know what it is doing. Falcon Guardian terminates it when either system says it should.

Krishna Gade framed the convergence dryly: "As AI security and observability converge, enterprises need to understand their agents' actions." Brad Palm of Dreadnode was more direct about the red-team side: "The people best equipped to find those failures think like attackers."

The market this deal signals is also worth naming. CrowdStrike is not alone in reading enterprise adoption of agents faster than enterprise governance of agents as a real gap. Palo Alto Networks, Cisco, and SentinelOne all have AI-agent-security products in some stage of launch. CrowdStrike is the first of the big four to put real dollars behind an acquisition pair in the space. The next analyst note sits under a different ticker.

What's interesting about the structure is that CrowdStrike paid roughly four times more for the observer than for the attacker. Either visibility is a bigger enterprise market than adversarial testing, or Fiddler's seven-year head start and nine-figure prior raise priced it where it was priced and nobody at CrowdStrike pushed back. Probably both.

What to watch

  1. The CrowdStrike Q3 FY27 earnings call (scheduled late November). Investors will press on the integration timeline, the Falcon Observe GA date, and whether the Kill Switch sits behind a higher pricing tier or inside existing Falcon Guardian subscriptions. The answer to that last question tells you whether CrowdStrike is monetising the feature or using it as a competitive moat.

  2. The competitive response. If Palo Alto Networks or SentinelOne announces an AI-agent-observability acquisition before the end of CrowdStrike's fiscal Q1 2028, the "secure your agents" category becomes a four-way bidding war for the remaining independents — Credo AI, WhyLabs, Protect AI, Robust Intelligence. The prices Fiddler and Dreadnode just set are now the floor.

  3. Whether the Kill Switch gets used. The marketing-vs-reality gap on big-red-button features is wide. CrowdStrike will not publish how often customers trigger it, but the trade press will leak the first incident where it gets pulled on a production agent. That is the day the feature stops being a keynote slide and starts being an operational artefact — and the day some CISO has to explain, in daylight, why they shut down a system at 3 a.m.

  4. Dreadnode's test harness, post-acquisition. The open question is whether CrowdStrike keeps publishing Dreadnode's red-team methods or walls them inside the Falcon Guardian product. Option one keeps the research community's goodwill. Option two is how big vendors usually behave with research-driven startups they just bought. Precedent is not encouraging.

The Monday headline was that CrowdStrike paid $495 million for two AI startups. The feature announcement underneath it is the more interesting story: a security vendor the Fortune 500 already trusts has decided the correct answer to agentic AI is a switch labelled OFF, mounted outside the model, with a human finger on top. That is a product philosophy. Whether it's the right one gets argued out over the next twelve months. On November 2, 2026, CrowdStrike put $495 million behind an answer.

* * *

Thanks for reading. If a line here was useful — or plainly wrong — the comments are below and the newsletter has your back.

Elsewhere in this issue

3 more
  1. 01

    News

    Anthropic's day one as ANT — opened $111.50, closed $97.17, founders cashed $1.96 billion

    Nov 20, 2026

  2. 02

    The Patch

    The Patch — November 20, 2026

    Nov 20, 2026

  3. 03

    News

    Anthropic priced its IPO above the range — $83 a share, 980 million shares, and ANT starts trading today

    Nov 19, 2026

Letters

Arguments, corrections, questions. Anonymous comments allowed; be kind, be specific.