Paintings Subscribe Tiếng Việt

22 November 2026

AI News/CrowdStrike pays about $495 million to see what AI agents are doing

CrowdStrike is buying Fiddler AI (agent observability) and Dreadnode (attacking AI before release). As agents spread through companies, the big question is no longer what an agent can do, but who can see what it is doing.

This month the security company CrowdStrike announced it is buying two AI startups at once. According to the Wall Street Journal, the deals total about $495 million (roughly $400 million for Fiddler AI and $95 million for Dreadnode). CrowdStrike did not disclose prices.

What do they do?

  • Fiddler AI does observability for AI agents: what an agent is doing, why it made a decision, and when it starts to drift. Its technology will become a product called Falcon Observe.
  • Dreadnode does attack testing before release: it automatically attacks a company’s models, reasoning loops and AI applications to find weaknesses before attackers do.

At the same time CrowdStrike launched an “AI Kill Switch” that stops unauthorised agents and prevents them from running again.

Why this caught my attention

In Moving faster than we can understand I argued that with many agents, the bottleneck is no longer building. It is understanding, controlling and steering.

This deal shows the market has reached the same conclusion, and is willing to pay half a billion dollars for it.

Three years ago, security meant protecting servers, laptops and user accounts. Now companies run hundreds or thousands of agents that act on their own, call APIs and read data. Each agent is a new employee who never sleeps, and a new attack surface.

Three lessons for people building agents

  1. Log every action an agent takes. If you cannot answer “what did the agent do at 3 a.m.?”, it is not ready for production.
  2. Attack your own agents before release. Prompt injection, poisoned data, misused tools: find the holes before someone else does.
  3. Always have a stop button. An agent you cannot switch off immediately is a risk, however clever it is.

The more powerful agents become, the more valuable it is not the agent itself, but the ability to see and control it.

Source: CyberScoop