Decoy Font: Cloaking Text from Artificial Intelligence

Decoy Font is a specialized TTF (TrueType Font) engineered to mask the actual content of your typing from AI systems, effectively creating a visual "smoke screen."

🛠️ Technical Overview & Licensing

Decoy Font is available for use in commercial, personal, and client-facing projects entirely for free. Its design is rooted in the DejaVu Sans Mono typeface; users should consult the full license for specific terms.

How the Illusion Works

The font employs a concept known as spatial frequencies to embed two distinct characters within a single glyph space.

1. The Foreground: Consists of sharp, thin outlines that form the "decoy" text.
2. The Background: Consists of a blurred, low-frequency mass that forms the "hidden" message.

Mathematically, we can think of the visual output $V$ as a combination of high-frequency components $H$ (the decoy) and low-frequency components $L$ (the secret): $V = H_{foreground} + L_{background}$

When these layers are merged, the perceived text changes based on the viewer's perspective:

• Up Close: The eye (and AI) picks up the high-contrast outlines.
• From a Distance: The high-frequency details blur, and the low-frequency mass becomes dominant.

Tip: If the hidden message isn't appearing, try squinting your eyes or physically moving your screen further away.

🤖 AI vs. Human Perception

Most AI vision systems analyze images by examining pixels at a high resolution (up close). Consequently, when a screenshot of Decoy Font is fed into an LLM, the model prioritizes the clearly defined foreground outlines.

Performance Benchmarks

Even cutting-edge models struggle with this simple optical trick. In testing:

• ChatGPT (GPT-4o/Sol): Fails to identify the hidden message.
• Gemini 3.5 (with Thinking): Fails to identify the hidden message.

🎨 The Science of Hybrid Images

Decoy Font is an application of hybrid image theory, a well-documented phenomenon in optical illusions. A classic example is the image that appears as Albert Einstein from one distance and Marilyn Monroe from another.

By applying this to typography, Mixfont has created a tool to protect intellectual property (IP) and private data from OCR (Optical Character Recognition) and AI scrapers.

🛡️ The Anti-AI Initiative

As AI capabilities evolve, the need to protect human-centric information grows. Decoy Font is part of a broader exploration at Mixfont into "Anti-AI" typography.

Comparison: Decoy vs. Ghost

⚠️ Limitations & Future Potential

It is important to note that Decoy Font is a guaranteed security measure an initial layer of confusion.

Potential vulnerabilities include:

• Advanced agents with coding capabilities that can programmatically filter frequencies.
• Specific prompting that instructs an AI to look for hidden patterns.

Despite this, it remains a potent deterrent against casual scraping.

Future Roadmap

1. Language Expansion: Extending the font to character-based languages like Chinese, where uniform character shapes could make hiding messages even more effective.
2. Security Applications: Integrating the technology into CAPTCHA systems.
3. Privacy: Using it for discreet communication between individuals.
4. Benchmarking: Using the font as a test to measure the evolution of LLM visual intelligence.

# Conceptual logic for a frequency-based font filter
def decode_decoy_font(image):
    low_pass_filter = apply_gaussian_blur(image)
    return "Hidden Message Detected" if low_pass_filter else "Decoy Only"

The creator of Decoy Font is currently developing a frontier AI font generator and welcomes community feedback and ideas on how to further evolve the intersection of typography and artificial intelligence.