ShieldFont is an open-source project that protects website text from AI scrapers that ignore robots.txt. The mechanism is a special web font: a human in the browser sees normal text, while the scraper receives a grammatically correct but useless variant for training. The project has already been described in a Hackaday review and has a live demo, but the protection breaks screen readers.
What happened
ShieldFont is an open-source project on GitHub (isaqueseneda/shieldfont, created 2026-07-23, around 440 stars). Its mechanism: a dictionary-based algorithm replaces about a quarter of the words in the HTML code with decoy variants built on ligatures, while the special font ShieldFont Optik — created together with the Danish type foundry Playtype in six weights from Regular to Black — renders normal text in the browser. The project has a white paper, published on 2026-07-30, and a live demo service at shieldfont.org; its review was published on August 14, 2026.
Context
robots.txt is a standard in which a site declares which pages robots are allowed to read, but it is only a polite request, and scrapers for LLM training often ignore it. That is why the arsenal of countermeasures already includes "labyrinth" traps like Nepenthes and Cloudflare's offerings, which keep bots on endless dynamic pages. ShieldFont's difference is its implementation as a regular web font without server-side code. From the perspective of ML research, there is no new capability here: it is a data protection mechanism, i.e., data poisoning at the typography level, not a new model or method.
Why this matters for the industry
For the industry, ShieldFont is not a block, but an economic mechanism: it does not prevent parsing, but makes large-scale scraping for LLM training significantly more expensive, because the scraper has to run rendering and glyph matching, essentially OCR, on each page. Combined with traps like Nepenthes and Cloudflare's offerings, this gives publishers a concrete opt-out from scraping that bypasses robots.txt and a precedent where such protection is implemented as a regular web font. The category of "content protection from AI scraping" is beginning to take shape as an independent market: the economics change not through blocks, but through cost, and each page parsing becomes more expensive.
Why this matters for users
For the reader, this is a tangible example of the war between publishers and LLMs: by inserting text into the demo at shieldfont.org, you can see what a human reads and what a scraper will steal. But there is a sobering counterweight: the protection breaks accessibility. Screen readers, which read text from HTML, receive distorted decoy variants, and legitimate search indexers are similarly affected. This is the real cost of anti-scraping, which is often silent in tech news.
What is still unknown / limitations
Public evidence is qualitative: there is a white paper and a demo, but no evaluation against specific scrapers, no figures on the cost of bypassing, and no measurements of resilience. The project is at the demo stage — around 440 stars, 3 points and 0 comments on Hacker News, with no measurable business metrics. For a large site, this is an experiment, not a production-ready solution, and it is unknown whether the cost of rendering and OCR is high enough to actually stop industrial scrapers.
Sources
- Hackaday review: ShieldFont: Bludgeoning AI Scrapers That Disrespect Robots.txt
- ShieldFont — White Paper (shieldfont.org)
- GitHub repository isaqueseneda/shieldfont
Author
Look at AI, editorial team