Infrastructure

Out of the Kernel?

On 10 September Microsoft hosted the Windows Endpoint Security Ecosystem Summit in Redmond. Endpoint security vendors attended, CrowdStrike among them, along with government representatives from the US and Europe. It’s a direct response to the July outage I wrote about in Twenty-One Fields.

Microsoft published a summary on 12 September. Two themes stood out:

  1. Safe deployment practices (SDP). Gradual, staged rollouts of updates, with monitoring and the ability to pause or roll back, applied consistently across the ecosystem.
  2. More security capability outside kernel mode. Microsoft says customers and partners have asked for it, and it plans to build new platform capabilities so security tools can do more of their work in user mode.

Not everyone wants to leave the kernel. Some vendors argued strongly that kernel access must remain an option, because that’s where you can see, and stop, the most sophisticated threats.

The Trade-Off

This is a genuine engineering tension, and I don’t think either side is wrong:

  • Kernel mode gives deep visibility and tamper resistance. But if something breaks there, it takes the whole machine down with it.
  • User mode contains the damage when a component fails. But it needs the operating system to provide rich enough hooks and protection for security tools to stay effective.

On the Linux side, technologies like eBPF have pushed in a similar direction: run the observability and security logic in a sandboxed, verified way, rather than loading arbitrary kernel modules.

My view, for what it’s worth: where the code runs matters, but how changes reach that code matters more. Most of July’s damage came from the speed and scale of the rollout. Safe deployment can prevent outages in either mode.

More Information