Duo MFA
A second factor on the paths that matter - and a deliberate answer to the question that decides what MFA is actually worth: what happens when the MFA service itself is unreachable.
What it is & why I use it
Duo provides the multi-factor layer on privileged access in AutomationLab. Where a password alone is not enough - administrative and remote access paths - a second factor (a push, a one-time code) has to be satisfied before access is granted. It is the "something you have" on top of "something you know."
I use it because credentials leak; that is a when, not an if. MFA is the control that makes a stolen password insufficient on its own, so the paths most worth protecting do not fall to a single reused or phished secret.
How I use it in the lab
- On privileged paths, not everything. The second factor guards the high-value access - administrative and remote entry - where the cost of a compromise is highest.
- Layered with the edge and identity. MFA sits alongside the zero-trust edge and the identity provider, so protected access requires both a proven identity and a second factor, not one or the other.
- A deliberate unreachable-service policy. The behavior when the MFA service cannot be reached is a decision I made on purpose, not a default I inherited (see the gotcha).
- Phishing-resistant factors preferred. Where possible the stronger factors are favored over a simple approve-the-push, because a push a user can reflexively accept is a weaker link.
Architecture
In action
Lessons learned & gotchas
- Have a break-glass path, and test it. Fail-closed only works if there is a pre-planned, audited way in when MFA is down - one that does not route through the failed service. An untested break-glass is just a different outage.
- A push you can reflexively approve is a weak factor. Prompt-bombing preys on fatigue - hammer someone with pushes until they tap approve. Number-matching and phishing-resistant factors blunt it; a bare approve/deny push does not.
- MFA protects the path, not the secret behind it. It stops a stolen password from being enough - it does nothing for a credential that is used by machines with no human to prompt. Those belong behind scoped secrets and service identity, not MFA.
Impact
Duo makes a leaked password insufficient on the paths that matter most. The lasting lesson is that MFA's real strength is set by its failure behavior, not its happy path: decide fail-open versus fail-closed on purpose, build the break-glass the closed choice demands, and prefer factors that a tired human cannot be tricked into approving.
Code & further reading
Any published access-policy notes are scrubbed of hostnames, addresses, usernames, and enrollment detail before they go public.