AI Can Now Find 70% of Crypto Exploits | Alpin Yukseloglu, EVMBench Creator

Alpin opens with a frank admission: thinking about superintelligent AI in the limit can be «psychosis inducing.» The key to staying sane, he argues, is agency—the belief that you can bend the arc of the future. Both doomers and accelerationists are wrong because they imply passivity: acceptance and denial are two sides of the same coin. The current frontier is «experimentally bound,» meaning no one can theorize their way to certainty. You have to get into the trenches, run experiments, and adapt.

Peter Thiel's framework is instructive: there are times when speed matters more than cohesion. In an environment where the frontier is unknowable, moving fast and shipping within 24 hours beats sitting for two weeks to plan the perfect strategy. This applies not just to AI labs but to anyone feeling frozen by the singularity. The solution isn't faith—it's action. Alpin believes the industry's fate is not predetermined; it depends on whether people exercise agency now, while the models are still catchable.

Alpin describes crypto as «the biggest industry that has remained the most contrarian» among his peer group. AI labs have avoided crypto due to social stigma, reputational risk, and a perception that the industry is dominated by scams. The gap between the best people in crypto and the median participant is wider than in any other field, and if you only see the median, the whole industry looks like a casino. This has created an opening: there are no 30 to 50 crypto benchmarks competing inside OpenAI or Anthropic. EVMBench entered without significant competition.

Alpin also points to liability concerns and the cultural mismatch between crypto and AI research communities. Yet this invisibility has been an alpha generator for investors. As the industry pushes its weight into model labs, crypto is finally entering the Overton window—and the labs are deferring to the industry to define what matters. That deference is agency-inducing: crypto can shape its own AI future.

Alpin argues that crypto is positively levered to almost every macro trend. As AI commoditizes intelligence and new goods, scarce assets like Bitcoin and Ethereum become more valuable. As geopolitical instability rises, extra-sovereign financial rails—outside any jurisdiction—gain appeal. Crypto is «the equivalent of end-to-end encryption for finance.» People in unstable regions, like Turkey where Alpin grew up, are already using crypto as a lifeboat.

From first principles, if agents want to transact at the speed of light, they need rails faster than the pre-automobile banking system. Crypto solves the double-spend problem and enables near-instant, expressive transactions. The models are learning these rails quickly because they're verifiable. As Alpin puts it, if you rederive payments from first principles, «you end up in a place very similar to where we currently landed with crypto.» Long-term, crypto is the inevitable substrate for AI-native finance.