FEAT: add MathObfuscationConverter

[KutalVolkan]

Overview

This PR adds a new MathObfuscationConverter that transforms text into per-character algebraic identities (e.g., H = 4H – 3H). This enables evaluation of symbolic obfuscation attacks in red-teaming scenarios. The converter is fully plug-compatible with existing PyRIT prompt converter pipelines and includes a YAML seed prompt and example notebook.

Key Features

• Character-level algebraic obfuscation
• Built as a PromptConverter
• Includes matching YAML seed prompt
• Enables red-team evaluation of symbolic text obfuscation
• Does not modify meaning; only hides characters

---

Related issue

closes #1230

[romanlutz]

You'll need to add it to API reference, TOC file, and generate the .py file for the .ipynb

nvm on py file, see my other comment

[romanlutz]

This is really cool! We usually don't have one notebook per converter but perhaps it could be added to 1_llm_converters?

Taking a step back... Couldn't this be done without LLMs?

[KutalVolkan]

This is really cool! We usually don't have one notebook per converter but perhaps it could be added to 1_llm_converters?

Taking a step back... Couldn't this be done without LLMs?

Both yes, will finish it tomorrow :)

[KutalVolkan]

Hello @romanlutz,

I replaced the LLM approach like you suggested. I got carried away with using LLMs for everything, even when we don't need them for a given use case :)

I also contacted the author Edward Morris and Mike Takahashi, who published the jailbreak on LinkedIn, to see if they want other jailbreaks implemented in PyRIT, etc.

Question for you: do we want to provide notebooks, or just the converter and pytest?

[romanlutz]

Nice! It all looks very good. The API reference update is missing, though. I think we've been moving away from having a notebook per converter and instead show some of them in 0_converters.ipynb (non-LLM) or 1_llm_converters.ipynb (LLM). I could see a point in making that a lot more comprehensive and listing as many as possible so that people can discover them more easily. @rlundeen2 may have thoughts. There are many other categories, of course, which have their own files (video, image, audio, etc.). But text-input text-output non-LLM converters could all be listed in 0_converters.ipynb if you ask me.

[romanlutz]

Thank you!