CrowdStrike Reportedly Buys AI Agent Security Startups Fiddler AI and Dreadnode: What It Means for Founders

If you build products with AI agents (software that carries out tasks on its own for a person or a company), a security company has reportedly just shown what it will pay for two things: watching agents at work and testing them for weaknesses.

What happened

TechCrunch reports, citing The Wall Street Journal, that CrowdStrike is buying two startups.¹ According to that report, CrowdStrike is paying about $400 million for Fiddler AI, which monitors AI agents inside companies, and about $95 million for Dreadnode, which automates security testing for AI models and agents.¹ Security testing means attacking a system on purpose so its weak points can be found and fixed. TechCrunch's headline puts the total at nearly $500 million and describes the market for securing agents as taking off.¹

The figures are approximate as reported, and we have not seen the terms of either deal.

What it means for founders

• If you build tools that secure agents, a buyer is reportedly paying both for watching agents and for testing them.
• If you build products that use agents, customers may start asking how your agent is watched and tested, and being able to show it may help you win them. That is a hypothesis, not a finding.

The deal sits beside what we have covered: our security roundup reported Elastic's account of intruders using AI tools and SpyCloud's report of malware stealing AI tool logins,² and our startups roundup reported that the agent funding we covered that week went to startups that keep a person reviewing the work and guard company data.³

So the decision is this: is watching and testing your agents part of what you build from the start, or something you show a customer only after they ask?

Our perspective: a hypothesis we are testing

What follows is our perspective, drawn only from what we have published. It is where our research begins, not a verdict.

Three of our published patterns bear on this, each with its status mark:

• A Record of Every Action, with a Way to Undo It: keep a record of every action your agents take, and make every change you can undoable. Status: Running for recording actions; Planned for one log.⁴
• Failures Reach a Person: send every failure to the person who can act on it. Status: Early for checking results and sending failures to a person; Planned for heartbeats.⁵
• Steps Only a Person May Take: name the few steps no agent takes, such as granting access. Status: Running for access and administrator changes and for the exact text of messages; Early for requests to a person.⁶

Our research note on running a publication with agents, marked Early, describes each kind of routine work as meant to run on its own only after its record earns it, with a random sample reviewed every week and any correction sending that kind of work back.⁷

Our hypothesis is an extension of that note, not something it says: a founder could use the same records and checks to decide how much each agent may do on its own. The note describes a publication, not a startup, and none of our pieces measures what customers want or what buyers pay. Our research note on how student founders build companies with agents is Planned: it proposes a study of how student founders build companies this way.⁸

The strongest case against

One reported deal is thin evidence of a market. There is also a serious opposing argument: if large security companies buy these capabilities, monitoring and testing may reach founders as part of larger security products, leaving little reason for a startup to build or show its own, and a founder's scarce time may be better spent on the product itself. Nothing we have published tests this argument either way.

What would change our mind

We would revise our view if founders who sell agent products report that customers rarely ask how agents are watched or tested, or if our own patterns, as they move from Early to Running, show that records and checks miss the failures that matter most.

One step this week

Write one page a customer could read: what your agent may do on its own, what only a person may do, where its actions are recorded, and who hears about a failure. If you cannot write that page, you have found your first piece of work.

For more on founders building with agents, see our Startups section.

This is not legal, financial, or investment advice.