Everyone and anyone seems up in arms about the (not new) threat of AI and/or quantum computing breaking cryptography.
Following Justin Drake’s viral post, “bunker mode” is practically a household phrase, despite being only a week old, and every token touting quantum preparedness has seen double-digit gains.
While it’s paramount that we address these issues and face them head-on, much of the discussion fails to identify initiatives already working to apply AI as a defense against these very threats, initiatives like those spearheaded by Eigen Labs’ Yukon.
— (@)
I’ve written about Yukon, or what would become Yukon, before: the open-research challenges Eigen Labs has been running since this summer.
These follow a structure where a technical problem is opened to the public, and participants work to “solve” it using a mixture of their developer knowledge and agentic assistance. Each proposed solution is automatically tested, with successful improvements becoming the starting point for the next attempt, spurring a constant flywheel of iteration.
The first of these challenges grew directly out of the original quantum scare. In March, Google’s quantum research first spooked everyone by suggesting that breaking Bitcoin and Ethereum’s cryptography could require substantially less quantum computing power than previously thought. It published its resource estimates and a way to verify them, but withheld the optimized circuits themselves.
— (@)
In response, Eigen Labs developed ECDSA.fail, turning the core problem behind Google's research into an open challenge. Using Google's published verifier as a starting point, participants could design and improve their own circuits for a crucial step in a potential quantum attack on Bitcoin and Ethereum's cryptography.
Researchers and AI agents have since reduced the challenge’s combined quantum resource-cost benchmark by roughly 88% from its initial baseline.
Of course, the improvements remain theoretical, and no existing quantum computer has demonstrated an attack of this scale.
Here's the power of open research. Withholding the circuits didn't make the threat disappear. It just made it harder to prepare against it. Yukon’s bet is that defenders are better off learning how cheap an attack can get in public than finding out by surprise.
That bet only matters if the model works, and outside of quantum, it already has. Several of these challenges have produced results now deployed in production or incorporated into project codebases:
Enjoying this article?
Subscribe to Bankless or sign in
Lighter.fast: A challenge to speed up Lighter’s transaction-proving software yielded a roughly 10x improvement in benchmark tests. The first batch of these optimizations, deployed in August, has helped cut the exchange’s actual proving time in half.
MLX.fast: Across several challenges, developers and their agents made models like Qwen roughly 2x faster and Poolside AI’s Laguna 2.6x faster on Apple hardware in benchmark tests: improvements which have since been integrated into Eigen’s private inference network, Darkbloom.
SNARK.fast: A collaboration involving Succinct and the Ethereum Foundation sped up an important calculation used in post-quantum cryptographic proofs by roughly 3.5x in early benchmark tests, with Succinct also incorporating the optimizations into their codebase.
That structure has since been spun into Yukon, a platform for these challenges launched in August, where a growing number of its challenges now test how vulnerable cryptography really is, develop safer alternatives, and make them practical to use.
Take these four challenges, for instance:
HashSmash
Launched October 5 with Zcash, Shielded Labs, and Project Eleven, HashSmash challenges researchers and AI agents to find weaknesses in popular hash functions like SHA-256, which secures Bitcoin, alongside a series of others. It’s these kinds of functions which underpin Justin Drake’s preferred defenses against AI and quantum attacks, so it’s imperative they’re tested as thoroughly as possible. So far, accepted attacks have only targeted simplified versions of SHA-256, not the complete algorithm, with more ambitious submissions awaiting review. You can review the ongoing results here.
— (@)
sig.golf
Developed with the Ethereum Foundation, sig.golf challenges participants to optimize the hash-based digital signatures that could replace today’s quantum-vulnerable transaction signatures, making them smaller and cheaper to verify while mathematically proving their security. Its leading result has nearly halved a combined measure of signature size and verification cost.
Quantum Safe Bitcoin
StarkWare’s Quantum Safe Bitcoin challenge aims to reduce the enormous GPU computing costs of an experimental method for protecting Bitcoin transfers against quantum attacks without changing Bitcoin’s rules. Its first mainnet demonstration in August required roughly $320 in computing costs, but Yukon participants have since reduced that estimate to $59, an 82% drop. That cheaper version has yet to be demonstrated in another live transaction.
— (@)
better.codes
Lastly, better.codes, another Ethereum Foundation collaboration, uses AI agents to strengthen the mathematical proofs supporting post-quantum cryptography. So far, participants have raised the security level they can mathematically prove from 64 bits to roughly 68, toward a 128-bit target.
By testing the weaknesses in existing cryptography, improving potential replacements, and strengthening the proofs behind them, Yukon is tackling the threat from multiple angles.
And while AI may be accelerating the race to break cryptography, it's encouraging to see how it could also help us stay ahead of the threat.
For anyone looking to contribute, Yukon is open to new challenge proposals.