What's this about?
An argument against fonts designed to be unreadable by AI transcription: they don't achieve their goal, and they degrade accessibility for the humans who need to read the text.
What exactly are anti-AI fonts trying to do?
They're typefaces deliberately distorted so optical character recognition fails while humans can still read them — a CAPTCHA applied to body text, intended to stop scrapers from ingesting published writing. The argument against is twofold. Modern vision models handle distorted text about as well as people, so the barrier is illusory. And the collateral damage is not: screen readers break, text selection breaks, translation breaks, and readers with low vision or dyslexia are locked out. You have inconvenienced no machine and excluded actual humans.
Why is this worth noticing?
HN 119 points, 81 comments
- It's the same lesson as the watermark stripper: technical measures against a determined, well-resourced adversary tend to burden ordinary users most.
- Accessibility is the decisive argument. Whatever you think about scraping, deliberately making text harder to read has victims who aren't the intended target.
- It fits the week's pattern of anti-AI countermeasures being evaluated on effectiveness rather than sentiment — alongside detection, watermarking and provenance.