Guest Post on False Marketing and PR Blitzes by Anthropic
By Guest Post
I've been keeping up with the sites and came across two news items related to Anthropic that I thought might be worth more attention.
The first involves cURL, the C library for transferring data over the Internet. Anthropic recently ran their "Mythos" model — the same one Wall Street claims will displace every Cybersecurity-Professional and Computer-Scientist alive — against the cURL codebase. On the 11th, cURL's founder and lead developer Daniel Stenberg reported back on the findings: FIVE supposed vulnerabilities flagged. Sounds alarming at first, right? Especially coming from the same person who once quipped that the "I" in LLM stood for intelligence a few years ago.
So what did this revolutionary "intelligent" Model actually find? Three results were simply errors in API documentation. The fourth was a genuine bug — but not a security vulnerability. The fifth and final finding was a real vulnerability, albeit an extraordinarily minor one. In Stenberg's own words: "The single confirmed vulnerability is going to end up a severity low CVE planned to get published in sync with our pending next curl release 8.21.0 in late June. The flaw is not going to make anyone grasp for breath." He went further, concluding that Mythos was "not particularly dangerous" and that the hype surrounding it was "primarily marketing." His full assessment: the model finds issues at no appreciably higher degree than tools that came before it — and even if it is marginally better, it isn't better enough to make a meaningful dent.
The second story is arguably more frustrating, and it involves the Zig programming language. I've never written anything in Zig myself, but the project occupies an interesting space — think of it as the pragmatic evolution of C rather than a wholesale replacement like Rust. It's entirely community-driven, with a strict no-AI/LLM policy baked into its contribution guidelines. Anthropic, apparently very bothered by that policy, forked the project a few weeks ago and published a flashy blog post claiming their LLM-assisted fork achieved a 4x compiler speedup. Impressive, if true - maybe LLMs were finally doing something genuinely useful? But as it turns out, their improved compiler only successfully compiles about 70% of the time. We have now officially entered the era of slop programming languages - not just languages with AI-generated slop written in them, but languages where the entire toolchain is slop. Would you trust a parachute with a 30% failure rate, or would you throw it aside and reach for something that actually works when you need it?
A programming language with a nondeterministic compiler feels like a throwback to the 1940s in the worst possible way. And knowing that this sort of thing is already making its way into tools like Bun - which Anthropic bought in December 2025, and which I use almost every day — makes me seriously want to start migrating to something freer, more trustworthy, and more aligned with our values, like Deno. At this point, I’m inclined to start encouraging others at Georgia Tech to ditch Bun as well.
This brings me back to Anthropic. The company founded by ex-OpenAI employees who were so alarmed by Sam Altman's reckless approach to AI safety that they walked out and started over — from scratch, on principle. Here we are a few years later, and they've grown larger and arguably more influential than the company they fled, while proving themselves just as committed to the black box and just as allergic to transparency. The organization built around protecting humanity from dangerous AI is now forking open-source/free-software projects in violation of their maintainers' explicit policies, because they couldn't get unlimited access to slop up someone else's codebase. Truly a gift to the planet - and an excellent example of why you should use Copyleft for big projects like a programming language, not just the super permissive but acceptable Expat license.
I'm not entirely sure what changed, but looking back to late 2022, when ChatGPT first grabbed the public's attention, it feels like the same crowd that had been busy cooking the planet with crypto mining rigs simply pivoted overnight to AI. Claude Code may well have been the bridge that got them there. Has Zig been in development for a long time without a non-beta release? Yes. Did Anthropic help at all with the development of Zig? No.
A lot of people my age are just tired of the nonsense. And when older people, armed with nothing but blind trust in these companies, try to lecture everyone else about how transformative all of this is, they're increasingly getting shut down on the spot. Just a few days ago, someone at the University of Central Florida tried to argue that AI was "the new Industrial Revolution" and got booed off the stage while pleading with the crowd to let her continue. Honestly, the video is worth watching for the laughs alone since it was their graduation. █