Everybody's Lost Their Minds
\nOpinions, mostly my own, on the importance of being and other things.\n\n\n [homepage] \n [blog] \n [jschauma@netmeister.org] \n [[@jschauma](https://mstdn.social/@jschauma/)] \n [RSS]September 16th, 2026
\n Men. Some would rather vomit up a rambling blog post\n wall of text that nobody's going to read than go to\n therapy. So here we are.\n \n\n \n\n I've seen my share of stupid over the years, but now\n people with no engineering background have started\n pitching "industry changing" solutions they cooked up\n in their agent infested homelab; people's emails\n read like bozotic LinkedIn-fluencer posts with punchy\n "it's not this, it's that" single-sentence\n paragraphs; online articles suffer a similar fate in\n their own convergence on Meh; and half of the people\n you interact with have turned into [meat proxies](https://sfisms.org/meat-proxy).\n \n\n \n\n Spending upwards of 75% of my time directly or\n indirectly dealing with AI every day has absolutely\n robbed me of most of my enjoyment of my work. Most\n days feel like that Twilight Zone where you wake up\n and you're the same, but everyone else is different.\n ([They were all\n like that.](https://youtu.be/mV8WzFa2COQ)) The cyber hype train has been going "choo choo" for a\n while, with the main AI companies trying to one-up\n each other [committing\n crimes](https://www.felonybench.com/) and somehow we let them; the conscious\n choice of anthropomorphic language by the companies is\n adapted unquestioned by the media, thereby absolving\n AI companies of their incompetence to secure their\n programs.\n \n\n \n\n Built on unapologetic exploitation of intellectual\n property and concentrating power in the hands of a\n very small number of US companies and oligarchs, these\n AI models not only lend themselves to [generation\n of Child Sexual Abuse Material](https://www.theguardian.com/technology/2026/jan/02/elon-musk-grok-ai-children-photos)—a product\n feature for [logged-in\n users](https://arstechnica.com/tech-policy/2026/01/musk-still-defending-groks-partial-nudes-as-california-ag-opens-probe/)—1but our continued use of\n them also directly supports their role in, e.g.,\n military target selection, such as [elementary\n schools](https://en.wikipedia.org/wiki/2026_Minab_school_attack).\n \n\n \n\n Meanwhile, every single company is happy to "ethics\n aside..." all of that and spend unimaginable amounts\n of "tokens"—a made-up currency following the\n casino model2—while staring at you\n blankly when you ask whether anybody has bothered to\n check if that support chatbot you vibe coded and which\n you fed all of your very mediocre at best\n "documentation" has any ROI.3 "Frontier Models" and AI-assisted vulnerability\n research is another topic with questionable results.\n Anthropic and OpenAI tried to one-up each other with\n how dangerous their models are and everybody who\n considers themselves an industry leader is now part of\n some mysteriously named "project" (like [Glasswing](https://www.anthropic.com/glasswing)\n and [Daybreak](https://openai.com/daybreak/), or\n [Athena](https://www.chainguard.dev/athena)\n and [Akrites](https://akrites.org/)) or\n co-signed various open letters (like [this](https://openai.com/collective-cyberdefense/)\n or [this](https://www.microsoft.com/en-us/corporate-responsibility/topics/open-weight/))\n to signal just how much they're totally not left out.\n \n\n \n\n Every participant in these project has thrown\n absolutely incredible amounts of engineering resources\n at the FOMO-induced, time-limited, "the first one's\n free" offer from Anthropic and OpenAI. Dozens of\n highly-paid security engineers had all of their\n priorities shifted and spent all of their\n time on this; the cost of the engineering hours\n spent on developing and adjusting AI vulnerability\n discovery harnesses, building new processes and\n pipelines to shoehorn thousands of findings into their\n vulnerability management processes, and of course\n working with the product owners on assessing and\n fixing the findings... all that must run in the many,\n many millions of dollars for each organization.\n \n\n \n\n And yet, despite having found literally\n thousands of new vulnerabilities (only [a\n fraction of which were reported to Open Source\n projects](https://www.vulncheck.com/blog/anthropic-glasswing-receipts), by the way), I don't think that we're\n any safer than before. That's because\n finding vulnerabilities has never been the\n bottleneck in information security. The bottleneck\n isn't even verifying a vulnerability report\n and validating its severity, as time consuming as that\n is. The bottleneck isn't determining the fix,\n creating the patch, or publishing a new release. The\n bottleneck is still, as ever before, getting the\n goddamn packages updated. Patching is still hard.\n \n\n \n\n Now imagine that we had spent all these resources on\n doing the basics: ensuring your organization has an\n up-to-date and comprehensive asset inventory with\n fine-grained package listings; building infrastructure\n that supports regular, frequent, and automated OS and\n applіcation updates; automatically rebooting systems\n when they hit, say, 30 days of uptime to ensure these\n updates are picked up; establishing comprehensive\n attack surface enumeration across all your IP space\n as well as all your cloud providers (what a\n concept!); the list of basic, fundamental defenses\n that nobody seems to actually do well goes on. Having\n a few dozen senior engineers dedicated for 6 months to\n overhauling all that, focusing on making\n patching easier, would, in my book, have been\n a much better investment, but that's just not very\n cyber at all.\n \n\n \n\n No matter what AI promises, human resources are still\n a zero-sum game, and every individual feeling super\n busy in their agentic silo doing a thousand things at\n once does not, in the end, help solve the kinds of\n projects that require cross-functional collaboration\n and team work. At the same time, AI companies are falling over\n themselves once again facetiously calling for their\n own regulation because, you know, they could\n accidentally end all mankind.\n \n\n \n\n If you actually thought your product will kill all\n humans, then you could, you know, like, just stop\n building the torment nexus. All by yourself, no\n government regulations required. Nobody's forcing you\n to play "Theaterwide Biotoxic and Chemical Warfare" or\n "Global Thermonuclear War". I mean, except your\n future shareholders and your greed. Alas, that\n wouldn't cockblock your competition...\n \n\n \n\n But you don't need to imagine AI destroying all\n humankind within the next few years via some sort of\n Skynet or Paperclip Maximizer scenario when in reality\n AI has of course already been hard\n at work here. The environmental impact of these\n companies is [staggering](https://news.bloomberglaw.com/environment-and-energy/us-data-centers-set-to-burn-more-natural-gas-than-most-nations).\n The AI race demands more and more water wasting, air\n polluting, fossil fuel powered data centers that\n absolutely nobody wants to live close to, and\n governments lift any and all environmental regulations\n for these companies who not too long ago at least\n pretended to have even the feeblest\n greenwashing commitments to carbon neutrality or\n renewable energy sources.\n \n\n \n\n But "[the\n world looks different now](https://www.nytimes.com/2026/08/08/climate/amazon-data-center-texas-pollution.html)", to which I can only say\n "No fucking shit, Sherlock. [IT'S\n ON FUCKING FIRE.](https://static01.nyt.com/images/2026/07/24/multimedia/24int-spain-france-fires-mpkz/24int-spain-france-fires-mpkz-superJumbo.jpg) Because of you." You know there are two ways for AI to achieve\n superhuman intelligence, right? One (theoretical) way\n is the mystical "recursive self-improvement" by AI.\n The other one is the path we're very clearly on: The\n agentic brain worms have been spreading, and it looks\n increasingly like everybody's lost their goddamn minds\n already. AI is the tool that dulls its users; it\n incrementally replaces understanding with a\n new dependency and addiction as you actively de-skill\n yourself.\n \n\n \n\n AI helps people find more vulnerabilities in existing\n code. To address those vulnerabilities, people use AI\n to generate patches. The resulting pull requests are\n then "reviewed" by AI. That is, the more AI is in the\n loop, the less we understand the code base. The\n mystical "human in the loop" often is nothing more\n than a rubber\n stamp.\n \n\n \n\n So what happens when things go bump? Complex systems\n fail in complex ways, and debugging code is an order\n of magnitude [harder\n than writing code](https://www.laws-of-software.com/laws/kernighan/). Debugging somebody else's code\n is harder still. Trying to debug large, complex,\n distributed systems consisting of components that are\n effectively opaque to your entire organization is\n going to be impossible.\n \n\n \n\n So no, I'm not going to set ethics aside and then\n actively offer myself up as tribute to self-amputate\n my brain. I'm sorry if everybody else can't get rid\n of the brain slugs, but at this point I'm just looking\n to get off this ride.September 16th, 2026
\n P.S.: And no, Claude is [not\n conscious](https://www.theatlantic.com/philosophy/2026/06/no-artificial-intelligence-is-not-conscious/687378/?gift=qrZWaSGx876y4WTlWSgdyE2DRbvX5US2rO-FfZV-AcQ&). J-Space my ass.Footnotes:
\n [1] Damn straight I use an emdash. Fuck\n you for devaluing it.\n ↩\n \n \n\n \n\n \n [2] "Results showed that participants\n gambled significantly more with chips than with real\n cash." [[citation provided](https://search.informit.org/doi/abs/10.3316/informit.223149445402224)]\n ↩\n \n \n\n \n\n \n [3] It doesnt: LLMs are Garbage-In/Garbage-Out—if\n you have shitty docs, the AI can at best polish that\n turd and still only spit out nothing of use.\n ↩Links:
\n ← [Sites using PQC (September 2026)]\n\n\n\n [homepage] \n [blog] \n [jschauma@netmeister.org] \n [[@jschauma](https://mstdn.social/@jschauma/)] \n [RSS]