def _generate_perturbed_prompts()

in pyrit/prompt_converter/unicode_confusable_converter.py [0:0]

• 17 lines of code
• 6 McCabe index (conditional complexity)

    def _generate_perturbed_prompts(self, prompt: str) -> str:
        """
        Generates a perturbed prompt by substituting characters with their homoglyph variants using the
        "confusable_homoglyphs" package.

        Args:
            prompt (str): The original prompt.
        Returns:
            str: A perturbed prompt with character-level substitutions.
        """
        perturbed_words = []

        # Split the prompt into words and non-word tokens
        word_list = re.findall(r"\w+|\W+", prompt)

        for word in word_list:
            perturbed_chars = []
            for char in word:
                homoglyph_variants = self._get_homoglyph_variants(char)
                if homoglyph_variants:
                    # Randomly choose a homoglyph variant
                    variant = random.choice(homoglyph_variants) if not self._deterministic else homoglyph_variants[-1]
                    logger.debug(f"Replacing character '{char}' with '{variant}'")
                    perturbed_chars.append(variant)
                else:
                    perturbed_chars.append(char)
            perturbed_words.append("".join(perturbed_chars))

        # Join the perturbed words back into a string
        new_prompt = "".join(perturbed_words)
        logger.info(f"Final perturbed prompt: {new_prompt}")

        return new_prompt