CrowdStrike has set out an unusually broad AI-agent security strategy, combining new Falcon Guardian controls with planned acquisitions of Fiddler AI and Dreadnode. Guardian adds an AI kill switch, a generally available gateway, agent-level hooks and supply-chain checks for malicious skills and Model Context Protocol servers. CrowdStrike says network inspection for Windows is available, while Mac support and further identity and cloud controls remain earlier in rollout.
Fiddler is intended to add pre-deployment evaluation, production traceability and behavioural accountability. Dreadnode would add automated tests across models, applications, agent reasoning loops and MCP servers, including jailbreak and prompt-injection techniques. Both transactions are subject to closing conditions. CrowdStrike says the resulting Fiddler-based Falcon Observe and Dreadnode testing capabilities are planned for general availability in the fourth quarter of its 2027 financial year.
The architecture is compelling because it connects release testing, runtime intervention and post-event explanation. It also risks collapsing several different assurance questions into one vendor control plane. Buyers should ask which evidence is independently verifiable, whether a kill switch works when an agent uses local models or unmonitored tools, and how policy recommendations move from tests into production. Claims such as a 98 per cent reduction in evaluation cost are vendor claims and need independent validation.
