--- title: Sorry Robot, Nice Try: Can AI Really Not Read This Font? description: A security researcher tests the viral claim that a new font is unreadable by AI. Initial tests confirm the AI fails to read blurred hidden text due to how... canonical: https://daily.dev/posts/sorry-robot-nice-try-can-ai-really-not-read-this-font--yibwkizaq twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Sorry Robot, Nice Try: Can AI Really Not Read This Font? | daily.dev og:description: A security researcher tests the viral claim that a new font is unreadable by AI. Initial tests confirm the AI fails to read blurred hidden text due to how... og:url: https://daily.dev/posts/sorry-robot-nice-try-can-ai-really-not-read-this-font--yibwkizaq og:image: https://api.daily.dev/og/posts/YiBwkIzAq.png og:image:alt: Sorry Robot, Nice Try: Can AI Really Not Read This Font? og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Sorry Robot, Nice Try: Can AI Really Not Read This Font? **[Medium](https://daily.dev/sources/medium_js)** · 5 min read · 1 upvotes · 0 comments ## Summary A security researcher tests the viral claim that a new font is unreadable by AI. Initial tests confirm the AI fails to read blurred hidden text due to how vision-language models encode images into lossy visual tokens. However, by instructing the model to actively process the image rather than passively read it — using median filtering, background removal, thresholding, and OCR — the hidden text 'HAPPY HUMAN' is successfully recovered. The key insight: 'AI cannot do X' usually means 'AI cannot do X passively,' and the gap between passive and tool-assisted capability can be bridged with a single instruction, not a new model. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### Why can't AI vision models read blurred text even though humans can see it clearly? Vision-language models divide images into patches processed through a lossy vision encoder that converts pixels into feature vectors before the language model ever sees them. If blurred letters are encoded only as a soft dark region, the letter identity never reaches the part of the model that generates answers, so no amount of re-prompting to 'look closer' helps since the detail is already gone from the input. _Anyone debugging why a model misreads an image can compare notes on encoder limits via daily.dev._ ### How do you make an AI model read text hidden by an anti-AI font trick? Instead of asking the model to look harder, instruct it to actively process the image rather than passively view it. Applying a median filter removes thin striped strokes while preserving the wide blurred shadow text, then background removal, thresholding, and optional deconvolution followed by OCR reliably reveals the hidden message, confirmed independently across multiple processing attempts. _Developers exploring AI image-processing workarounds can track similar technique writeups on daily.dev._ ## Similar posts on daily.dev - [Ghost Font: The Anti-AI Font Only Humans Can Read](https://daily.dev/posts/ghost-font-the-anti-ai-font-only-humans-can-read-byqnplkwl) · Hacker News · 1 upvotes · 0 comments - [AI detection tools cannot prove that text is AI-generated](https://daily.dev/posts/ai-detection-tools-cannot-prove-that-text-is-ai-generated-fvd95pteg) · sean goedecke · 2 upvotes · 0 comments - [What happens when you put AI to work deciphering lost languages?](https://daily.dev/posts/what-happens-when-you-put-ai-to-work-deciphering-lost-languages--shhgx6xjn) · Ars Technica · 0 upvotes · 0 comments --- Tags: [#machine-learning](https://daily.dev/tags/machine-learning), [#llm](https://daily.dev/tags/llm), [#computer-vision](https://daily.dev/tags/computer-vision), [#image-processing](https://daily.dev/tags/image-processing) [View this post on daily.dev](https://daily.dev/posts/sorry-robot-nice-try-can-ai-really-not-read-this-font--yibwkizaq) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"Sorry Robot, Nice Try: Can AI Really Not Read This Font?","url":"https://daily.dev/posts/sorry-robot-nice-try-can-ai-really-not-read-this-font--yibwkizaq","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/sorry-robot-nice-try-can-ai-really-not-read-this-font--yibwkizaq"},"datePublished":"2026-07-21T20:07:15.519Z","dateModified":"2026-09-14T08:02:18.486Z","description":"A security researcher tests the viral claim that a new font is unreadable by AI. Initial tests confirm the AI fails to read blurred hidden text due to how...","image":"https://media.daily.dev/image/upload/s--foaA6JGU--/f_auto/v1722860399/public/Placeholder%2004","thumbnailUrl":"https://media.daily.dev/image/upload/s--foaA6JGU--/f_auto/v1722860399/public/Placeholder%2004","isAccessibleForFree":true,"articleSection":"Medium","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Medium","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/medium","url":"https://daily.dev/sources/medium_js"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/sorry-robot-nice-try-can-ai-really-not-read-this-font--yibwkizaq","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":1},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"machine-learning,llm,computer-vision,image-processing","timeRequired":"PT5M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Medium","item":"https://daily.dev/sources/medium_js"},{"@type":"ListItem","position":3,"name":"Sorry Robot, Nice Try: Can AI Really Not Read This Font?"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/sorry-robot-nice-try-can-ai-really-not-read-this-font--yibwkizaq#faq","mainEntity":[{"@type":"Question","name":"Why can't AI vision models read blurred text even though humans can see it clearly?","acceptedAnswer":{"@type":"Answer","text":"Vision-language models divide images into patches processed through a lossy vision encoder that converts pixels into feature vectors before the language model ever sees them. If blurred letters are encoded only as a soft dark region, the letter identity never reaches the part of the model that generates answers, so no amount of re-prompting to 'look closer' helps since the detail is already gone from the input. Anyone debugging why a model misreads an image can compare notes on encoder limits via daily.dev."}},{"@type":"Question","name":"How do you make an AI model read text hidden by an anti-AI font trick?","acceptedAnswer":{"@type":"Answer","text":"Instead of asking the model to look harder, instruct it to actively process the image rather than passively view it. Applying a median filter removes thin striped strokes while preserving the wide blurred shadow text, then background removal, thresholding, and optional deconvolution followed by OCR reliably reveals the hidden message, confirmed independently across multiple processing attempts. Developers exploring AI image-processing workarounds can track similar technique writeups on daily.dev."}}]} ```