Everyone seems to be up in arms about the (not new) threat of AI and/or quantum computing breaking cryptography.
Following Justin Drake viral message“bunker mode” is practically a household phrase, even though it’s only a week old, and every token touting quantum readiness has seen double-digit gains.
While it is critical that we address these issues and confront them head on, much of the discussion fails to identify already effective initiatives to apply AI as a defense against these same threats, initiatives like those led by Eigen Labs. Yukon.
— (@)
I have already written about the Yukon, or what would become the Yukon: the open research challenges that Eigen Labs has been launching since this summer.
These follow a structure in which a technical issue is open to the public and participants work to “solve” it using a mix of their developer knowledge and assistance from an agent. Each proposed solution is automatically tested, with successful improvements becoming the starting point for the next attempt, stimulating a constant flywheel of iterations.
The first of these challenges arose directly from the initial quantum fear. In March, Google’s quantum search initially scared everyone by suggesting that breaking the cryptography of Bitcoin and Ethereum might require far less quantum computing power than previously thought. It has published its resource estimates and a way to verify them, but retained the optimized circuits themselves.
— (@)
In response, Eigen Labs developed ECDSA.failturning Google’s core search problem into an open challenge. Using the checker released by Google as a starting point, participants were able to design and improve their own circuits for a crucial step in a potential quantum attack on Bitcoin and Ethereum cryptography.
Researchers and AI agents have since reduced the combined quantum resource cost benchmark of the Challenge by approximately 88% from its initial baseline.
Of course, improvements remain theoretical and no existing quantum computer has demonstrated an attack of this magnitude.
This is the power of open inquiry. Maintaining the circuits has not made the threat disappear. It just made it harder to prepare. Yukon’s bet is that defenders are better off learning in public how cheap an attack can be rather than finding out by surprise.
This bet only matters if the model works, and outside of quantum, it already does. Several of these challenges produced results now deployed in production or integrated into the project codebases:
Do you like this article?
Subscribe to Bankless or log in
-
Lighter, faster: A challenge to accelerate Lighter’s proof of transactions software resulted in an approximately 10x improvement in benchmark tests. The first batch of these optimizations, rolled out in August, cut the exchange’s real-world testing time in half.
-
MLX.rapid: Faced with several challenges, developers and their agents have created models like Qwen about 2x faster and those of Poolside AI. Laguna 2.6x faster on Apple hardware in benchmark tests: improvements that have since been integrated into Eigen’s private inference network, Darkbloom.
-
SNARK.fast: A collaboration involving Succinct and the Ethereum Foundation has accelerated an important computation used in post-quantum cryptographic proofs by approximately 3.5x in early benchmark testsSuccinct also including optimizations in their codebase.
This structure has since been transformed into Yukon, a platform for these challenges launched in August, where a growing number of its challenges now test the real vulnerability of cryptography, develop more secure alternatives and make them practical to use.
Take for example these four challenges:
HashSmash
Launched October 5 with Zcash, Shielded Labs and Project Eleven, HashSmash challenges researchers and AI agents to find weaknesses in popular hash functions such as SHA-256, which secures Bitcoin, alongside a range of others. These are the types of functions that underpin Justin Drake’s favorite defenses against AI and quantum attacks. It is therefore imperative that they are tested as thoroughly as possible. So far, accepted attacks have only targeted simplified versions of SHA-256, not the full algorithm, with more ambitious submissions awaiting review. You can view the current results here.
— (@)
sig.golf
Developed with the Ethereum Foundation, sig.golf challenges participants to optimize hash-based digital signatures that could replace today’s vulnerable quantum transaction signatures, making them smaller and cheaper to verify while mathematically proving their security. Its main result was to reduce a combined measure of signature size and verification cost by almost half.
Quantum Secure Bitcoin
StarkWare Quantum Safe Bitcoin Challenge aims to reduce the enormous GPU computing costs of an experimental method of protecting Bitcoin transfers against quantum attacks without changing Bitcoin’s rules. It is first mainnet demonstration in August it cost about $320 in IT costs, but Yukon participants have since reduced that estimate to $59, an 82% drop. This cheaper version has yet to be demonstrated in another live transaction.
— (@)
best.codes
Lately, best.codesanother Ethereum Foundation collaboration, uses AI agents to strengthen mathematical proofs supporting post-quantum cryptography. So far, participants have increased the level of security they can prove mathematically from 64 bits to around 68, reaching a goal of 128 bits.
By testing weaknesses in existing cryptography, improving potential replacements, and strengthening the evidence behind them, Yukon is tackling the threat from multiple angles.
And while AI may be accelerating the race to crack crypto, it’s encouraging to see how it could also help us stay ahead of the threat.
For anyone who would like to contribute, Yukon is open to new challenge proposals.
About the Author
