Lars Gyllenhaal
Writing
3 min read

On the False Economy of Complexity

The dominant theory of security failure holds that people are the problem. The user clicked the phishing link. The administrator reused the password. The developer shipped the unvalidated input. This theory is comfortable because it locates the failure outside the system — in the human element, which is famously difficult to fix and therefore provides indefinite cover for not fixing anything else.

It is also wrong. Or rather, it is correct in the way that blaming a flood on the water is correct. Technically accurate. Completely useless.

Systems that can be compromised by a single bad decision made by a single person are not systems with a people problem. They are systems with an architecture problem. The architecture chose to depend on that decision being correct every time. That is a design choice, and it is a bad one. Trust that must be earned continuously, without error, over an indefinitely long period by an indefinitely large number of people will eventually be broken. This is not a cynical observation about human nature. It is a statement about probability.

The organizations that spend the most on security awareness training tend to have the weakest architectures. This is not a coincidence. The training is performing the function of the architecture — poorly, unreliably, and at continuous cost — while the architecture remains unchanged. Every dollar spent teaching employees to recognize phishing attempts is partly a subsidy for an email infrastructure that cannot authenticate senders. Every hour spent on password policy is partly a subsidy for a system that cannot enforce meaningful session controls.

Complexity compounds this. A system no one fully understands cannot be defended by anyone. The threat model for a three-hundred-service microservices application where no single engineer can trace a request from ingress to data store is not a threat model at all. It is a document produced to satisfy a requirement. The actual security properties of that system are unknowable except adversarially — which is to say, by someone who is trying to find out by breaking it.

Simplicity is not always achievable. But the consistent preference for complexity — for the impressive architecture, the extensive toolchain, the layered controls that no one has time to configure correctly — is a preference with a cost that is rarely computed honestly. The cost is paid later, by someone else, in circumstances that are less convenient than a quarterly review.

What I find, almost without exception, is that the organizations with the most mature security postures are the ones that have made deliberate choices about what they trust and have verified those choices. Not the ones with the most tools. Not the ones with the largest security teams. The ones that built systems that reflect an honest accounting of their actual threat surface and do not require constant human correctness to remain intact.

That is not a complicated idea. It is, however, an unpopular one.