Invitation to Crack Codes Using AI
Could you use AI to crack a cypher? For instance, predict the next bits in bitstreams produced by a high-quality PRNG (pseudo-random generator). Or correctly guessing the next bit in sub-sequences of 100,000 consecutive binary digits of π starting at arbitrary positions, with a success rate above 55%. Without knowing that the digits come from π.
Beating random guesses by even a short margin renders a cyber unsecure — as they are supposed to be totally unpredictable if properly designed. So is the number π, supposedly. Likewise, having trading strategies with a bare but consistent 55% success rate makes the difference between becoming a billionaire or staying poor. A small success beyond luck opens the door towards further analyses that may eventually lead to fully cracking the code in question.
The challenge
In this article, I invite you to use deep neural networks or any other tools (AI-based or not) to crack my new cypher. In short, using the same tools that predict the next token to power large language models, but here to predict the next bits based on previous bits in bitstreams generated by my non-periodic random generator NPG. Even achieving a modest 50.1% success rate, as long as it is statistically significant, would be considered a success. It would defeat my claim that it is uncrackable.
I actually run several tests, including next-bit prediction using deep neural networks, but could not do better than a 50% correct prediction rate. By contrast, I was able to do much better when trying to crack other PRNGs. Including some widely used such as PCG64 which now replaces the Mersenne Twister in NumPy and other platforms. That said, I haven’t spent months trying to reverse-engineer my own system. My claim that it is uncrackable is based on limited testing, along with number theoretic arguments.
How to participate?
I describe my cypher in my previous post, here. It comes with bitstream data, as well as the Python code used in my tests, including deep neural networks. This is a starting point. I am willing to share more data, including the full PRNG state, with various bitstreams generated with different parameters, also sharing the parameters attached to each sequence to help you crack my system: get above 50% correct prediction for the next bit, on average and consistently. On some or all bitstreams. For more detail, contact me at vincent@bondingAI.io
What’s in it for you?
Gaining experience working on an enterprise-grade project related to cybersecurity, great references for your career, working with one of the top AI scientists in the world, playing with state-of-the-art AI, and even the possibility to get hired or compensated should this project leads to monetization on my side. The PRNG that produces these bits is much faster, stronger and more secure than any other one available on the market. To back this claim, I need your help. Any improvement, in case weaknesses are uncovered and fixed, would be a great success with your contribution prominently featured and shared on my large network and in specific cybersecurity, quant, and related circles. Unless you prefer anonymity for obvious reasons.
Get started today! Feel free to contact me by email or on LinkedIn, here. Get your free copy of my new book where it all started, here. See book review, here. And learn about the possibility to earn our new AI Cybersecurity Certification.
About the Author
Vincent Granville is a pioneering GenAI scientist, co-founder at BondingAI.io, the LLM 2.0 platform for hallucination-free, secure, in-house, lightning-fast Enterprise AI at scale with zero weight and no GPU. He is also author (Elsevier, Wiley), publisher, and successful entrepreneur with multi-million-dollar exit. Vincent’s past corporate experience includes Visa, Wells Fargo, eBay, NBC, Microsoft, and CNET. He completed a post-doc in computational statistics at University of Cambridge.
