The usual approach is to trust the security tool. Detect more. Patch faster. Ship the fix.

That fails when the failure includes the tool itself — a bad update, a hijacked path, a module that should not have had broad power. Recovery still matters. It is not the same as limiting damage while something is wrong.

We design for the second problem: assume a part can be compromised, and make lateral movement hard.

AL’S-OPS LLC is a Utah company building privacy-first infrastructure software.

What changed

  • Clear separation between watching and acting — detection does not automatically inherit high-impact power
  • Actions stay narrow and explicit; bigger steps wait for operator approval
  • Limits live with the server, not only in a remote policy the attacker might also reach
  • Integrity checks so a component that does not match what we shipped does not get a free pass to run powerful paths
  • Expiry and undo on automated steps so a bad decision decays instead of living forever

The real problem

If any module, agent path, or host-side piece can be wrong, “we’ll patch it” is incomplete. Patching is what you do after. Containment is what limits spread during.

A tool that can only succeed when every part is healthy is fragile under the conditions operators actually hit.

What we are not doing

  • Claiming compromise is impossible
  • Building a heavier platform that needs a security team to interpret
  • Hiding how controls behave when they are on
  • Treating “documented” as the same as “enforced on the host”

Assume a part can fail. Make sure that failure does not get the rest of the system for free.

Originally published on the Al's-Ops blog.