The Internet Optimized Away Its Own Backup — and Called It Efficiency
June 21, 2026 · Kevin Kelly, Out of Control~5 min read
In March 2026, one of the most-deployed pieces of software on the planet quietly stopped getting security patches. Ingress NGINX — the thing routing traffic into a large share of the world's Kubernetes clusters — was retired, not because a better tool replaced it, but because the one or two volunteers maintaining it on nights and weekends finally ran out. No villain, no breach, no dramatic exploit. Just a load-bearing wall that turned out to be held up by a single tired person, and a deadline after which nobody is holding it up at all.
A backbone you can count on one hand
This is not a freak event. It is the shape of the whole foundation. The XZ Utils backdoor of 2024 — the one that came within weeks of handing an attacker a master key to much of the internet — succeeded precisely because XZ was maintained by essentially one exhausted Finnish developer working unpaid in his spare time. The attacker did not break the code. They befriended the maintainer, waited out his burnout, and were handed commit access as a relief. A 2026 survey of open-source maintainers found that 60% are paid nothing for work the global economy runs on, and 58% have quit or seriously considered quitting. The software under your bank, your hospital, your phone is propped up by people you could fit in a small room, and a worrying number of them want out.
We mistook redundancy for waste
Here Kevin Kelly's Out of Control, written back in 1994, reads like a warning we filed and forgot. Kelly's whole argument is that anything alive and adaptive — an ecosystem, a market, the internet itself — survives because it is built the opposite way from a machine. Living systems are distributed across many small redundant units, most of them merely "good enough" rather than optimal. The slack, the duplication, the slightly inefficient overlap — that is not waste to be trimmed. It is the resilience. A desert survives drought because its rainfall is erratic and its species many; strip it down to one perfectly tuned crop and the first bad year ends it. For thirty years the software industry has done the trimming. We deduplicated, consolidated, and standardized our way down to one canonical library for each job — and felt clever doing it, because every removed copy looked like saved effort.
A living system survives because many redundant, merely-good-enough parts overlap: when one fails, others carry the load. The internet "optimized" that down to one canonical library per job, often held up by a single unpaid maintainer — so one burnout becomes a system-wide failure. Framework: Kevin Kelly, Out of Control. Real basis: the XZ Utils backdoor (2024) and the March 2026 retirement of Ingress NGINX; 2026 maintainer survey, ~60% unpaid. A reflection, not security advice.
Efficiency and resilience pull in opposite directions
The trap is that the optimization looks like progress at every step. One battle-tested library beats five half-built ones. Standardizing on the canonical tool removes confusion, saves engineer-hours, ships features faster. Each consolidation is locally correct, which is exactly why the industry kept doing it for thirty years without alarm. But Kelly's point is that a system tuned for peak efficiency is, by the same act, tuned against resilience. The two genuinely trade off. Every backup you delete because it was "never used" is gone the day you need it. We ran that experiment across the entire software supply chain, removed the slack as fast as we found it, and now act surprised that the structure is brittle. It is brittle because we made it efficient. Those were always the same decision.
Why the fix won't come from a hero
The instinct now is to look for someone to blame or someone to save us — a foundation, a corporate sponsor, a heroic new maintainer to parachute in. Useful, but it misreads the problem, because the problem is structural, not personal. A single well-funded maintainer for XZ would still be a single point of failure, just a better-rested one. Kelly's living systems don't survive on heroes; they survive on having many paths, so that no one path is load-bearing. Resilience is not a person you hire. It is slack you deliberately keep — second implementations, funded redundancy, the boring duplicate that exists precisely so it can sit unused until the year it saves you. The XZ attacker understood this better than the industry did. They didn't need to defeat the system. They just needed to find the one tired human the whole thing rested on.
What this means for you
You don't run the internet, but you run something — a service, a team, a stack of dependencies, a household budget. Go find your XZ: the one critical thing held up by one person, one vendor, one library, one login only you remember. The industry's mistake is yours in miniature every time you delete the backup that "wastes" space, cross-train nobody because it's slower, or route everything through the one teammate who never says no. Slack feels like inefficiency right up until the day it is the only thing standing between you and an outage. Keep some on purpose. The cheapest insurance you will ever buy is the duplicate you hope to never use.
We spent thirty years deleting redundancy and calling it efficiency. A living system without slack doesn't run lean — it runs until the first bad day, and then it stops.
Resilience was never a hero you hire. It's the boring backup you keep precisely because you hope to never need it.
Source: framework from Kevin Kelly, Out of Control (1994) — living, adaptive systems survive on distributed redundancy and "good enough" multiplicity, not single-point optimization. Real-world basis: the XZ Utils backdoor (CVE-2024-3094), the Kubernetes retirement of Ingress NGINX with best-effort maintenance ending March 2026, and a 2026 open-source maintainer survey finding ~60% unpaid and ~58% having quit or considered quitting. A reflection, not security or operational advice.