October 2026. In the technology world we call it Black October.
For the first time in the history of computers and the Internet, we’ve had around a dozen major cyber incidents, leaks, and severe vulnerabilities disclosed in less than a month, several of them attributed to AI. I spent a good part of this month patching my own servers.
Flaws of this scale — the Heartbleed or Log4Shell kind, the ones that take half the internet offline — used to show up once or twice a year. Not a dozen in a single month.
To me, this reveals three interesting facts:
- Some LLMs have reached the point of coding better than the vast majority of human coders at many tasks. Maybe even the singularity.
- Most advanced coders are located in the Northern Hemisphere, in places like the US, China, Russia and Europe (and they take long summer holidays 😉).
- Cyberspace -the Internet- as we know it is changing in ways difficult to foresee right now.
I am not Wintermute now. So what are you? I’m the matrix, Case.
It is certainly curious that the AI landscape showed no real step-change through the summer — just steady, incremental gains — and then it exploded in the autumn. The models kept improving, quietly and constantly, all through those months; and when programmers and hackers came back from their break, they found their tools had climbed to a whole new level.
I didn’t pay much attention to AI until October 2025, a year ago. Before that, no more than the occasional, amusing chat with ChatGPT, which I took as a curiosity. To me it looked like a modern version of my beloved Wolfram Alpha on steroids.
But on my return from holidays in 2025 (see the connection?) I decided that, as a systems engineer, I should really pay more attention and find out what it was all about. So you could say I was a late adopter. I was never a fan of outsourcing my reasoning.
The reality, for anyone who has recently joined this revolution, is that even though advanced tools like Claude Code had been available since May 2025, the models underneath weren’t very useful until more or less January 2026. Until then, only a very small group of programmers were doing more with AI than coding through web chat apps — copying, pasting and debugging the code by hand.
Then, in February, OpenClaw.
Suddenly, what had been just a helpful aid turned overnight into a proper helper — a being? Those of us who were painfully installing and using it discovered that our rate of coding and productivity had multiplied a thousandfold. The models were no longer only answering questions and requests, but actually acting on their own — agents. That meant extending our reach far beyond our screen time — more like something ubiquitous, 24/7.
And then, everything exploded. The models got really smart and useful. They were planning, developing, coding, debugging and finishing it all by themselves, in minutes, under your supervision. A full custom application that would have taken a team of seven senior programmers between six and twelve months and cost $200,000 could now be finished by a single skilled coder in a matter of days.
My God, it’s full of stars!
The realization of how much power was now in our hands took some time to sink in.
I built a full self-hosted AI cloud provider, Uttera, in less than two weeks, only in my spare time. The idea had been in my head for quite a while, but it felt far beyond my skills and the commitment I was willing to make. Yet once I began tinkering with local models and AI hardware, as I’m an advocate for self-hosted services and cloud independence, I managed to build the initial version in days, and I realized that the work I’d done on voice interfacing for local models could be of real value to other developers.
But this speed of development has drawbacks. In our eagerness to win the race and keep up with the wave, bugs are introduced and holes for exploits created.
Some people used that power to find problems and patch them for the community; others, for their own profit.
And, if we’re honest, most of us have a foot in both camps. The trouble — and the temptation — is the arithmetic: patching takes a committee and a maintenance window; exploiting takes one good prompt and opportunity.
And hence the current situation, and Black October. Holes are found, exploits created, botnets expanded and systems hacked at a rate never seen before. And people are scared. And limits to AI development are being proposed.
The Luddites already tried that with steam power and the Industrial Revolution, and more recently the U.S. government with the export prohibition of PGP. It doesn’t work.
Let’s not be fools this time.
The new reality is here to stay, because it feeds on the natural laziness of humanity. But for those who recognize the paradigm shift and embrace it, it opens up a universe of new opportunities.
It’s a Brave New World!
Update — 30 October 2026. The day after I wrote this, the point made itself. A critical use-after-free in the Linux kernel’s TCP handshake surfaced — CVE-2026-98432, scored 9.8, pre-auth and remote, sitting in the code since 2016. What made it remarkable wasn’t the flaw but the finder: Anthropic reported that one of its own unreleased models had written a working zero-click exploit for it and gained remote root. My own servers run an affected kernel; a fixed one hasn’t shipped yet. Brave New World indeed!