Artificial intelligence is supercharging hackers' ability to find software vulnerabilities faster than traditional security can patch them, according to NEAR Protocol co-founder Illia Polosukhin, who argues that developers should be turning to mathematical proofs to secure their code.
"I would actually say last October — I think Balancer was the first one — that’s when AI-assisted hacking started," Polosukhin said Monday in an interview with Gareth Jenkinson on The Block's The Starting Block podcast.
Polosukhin said cybersecurity up to this point has mostly involved "smart people" reviewing code and hoping that no one smarter finds something they missed. That mindset was already imperfect before the rise of AI, he said, but newer and more capable models are making it even less sustainable.
"We need a different way of thinking," Polosukhin said. "It’s not like, 'Oh, we’re going to review everything with a better model and hope that the next better model is not going to find it.'"
His proposed solution is formal verification that mathematically proves code "does what it says it does." This process has been historically expensive, slow, and labor-intensive, but Polosukhin said AI can now help generate most of the necessary proof, which can then be independently checked.
