CanaryTrap

CanaryTrap is a Python package that uses the concept of a "canary trap" to watermark text using unicode spaces. By converting regular spaces to unicode spaces according to a binary pattern, you can add a hidden signature to your text.

Installation

You can install CanaryTrap using pip:

pip install canarytrap

How to use

First, import the package:

import canarytrap as canary

To watermark your text, use the unicode_space_encode function. You can either provide a binary string or let the function generate a random one:

text = "Hello world this is a test"
watermarked_text, binary_str = canary.unicode_space_encode(text)
print(watermarked_text)  # Watermarked text
print(binary_str)  # Binary string used for watermarking

You can convert the watermarked text back to binary using the unicode_space_to_binary_str function:

binary_str_back = canary.unicode_space_to_binary_str(watermarked_text)
print(binary_str_back)  # Should be the same as binary_str

To check if a text matches a given binary string, use the unicode_space_match function:

match = canary.unicode_space_match(watermarked_text, binary_str)
print(match)  # Should be 1.0 if text is not modified

Example Scenario

You have someone who is leaking private documents and want to find out who. You're about to send an email to a bunch of different employees. Let's use CanaryTrap to find out who it is.

First, let's encode our email

import canarytrap as canary

staff_email = "Hello staff, this is a test email. Please do not share this email with anyone else."

watermarked_email, binary_str = canary.unicode_space_encode(staff_email)

with open("email.txt", "w", encoding="utf-8") as f:
    f.write(watermarked_email)

with open("binary.txt", "w", encoding="utf-8") as f:
    f.write(binary_str)

The unicode characters will not appear (Is an empty character). Try copying that text into a code editor to see it

Hello‎ staff,‎ this is a test email.‎ Please do‎ not share‎ this‎ email‎ with anyone‎ else.

You can then check some text against the watermark:

import canarytrap as canary

watermarked_text = "Hello‎ staff,‎ this is a test email.‎ Please do‎ not share‎ this‎ email‎ with anyone‎ else."

match = canary.text.unicode_space_match(watermarked_text, binary_str)
print(match)  # Should be 1.0 if text is not modified

Note

If the watermarked text is edited (spaces are added, removed, or replaced), the match percentage returned by the unicode_space_match function may be less than 1.0. If you get a match percentage of less than 1.0, it means that the text was likely edited after being watermarked. The match percentage gives an estimate of how much of the original watermark remains in the text.