Illia Polosukhin warns that AI-powered attacks on blockchain systems started last October and says the industry needs to fight fire with fire
Illia Polosukhin, the co-founder of NEAR Protocol and one of the researchers behind the landmark “Attention Is All You Need” paper that birthed the Transformer architecture powering today’s AI models, is sounding the alarm on a new class of threats to blockchain security. According to Polosukhin, AI-assisted hacking started emerging in October, and the crypto industry needs to respond by weaponizing the same technology for defense.
His prescription is straightforward: use AI to dramatically improve formal verification of smart contract code. In English, that means deploying machine learning to mathematically prove that code does what it’s supposed to do, and nothing else, before it ever touches a mainnet.
The threat landscape is shifting fast #
The implications for blockchain are uniquely severe. Smart contracts are immutable once deployed. A vulnerability exploited by an AI scanning tool doesn’t just mean a data breach, it means funds are gone. No rollback, no customer service hotline, no insurance claim.
Traditional security approaches in crypto have leaned heavily on code audits, bug bounties, and what Polosukhin characterizes as probabilistic defenses. Rather than simply training systems against known attack vectors like prompt injection, he advocates treating every user input as a potential threat vector by default.
Fighting AI with AI #
Formal verification uses mathematical proofs to guarantee that code behaves exactly as intended under all possible conditions. AI could change that calculus by assisting with or even automating portions of the formal verification process, reducing the time required significantly.
NEAR AI, the protocol’s dedicated artificial intelligence arm, is actively building in this direction. The initiative encompasses confidential inference, private AI computing, and what NEAR describes as agent economy standards. One notable project in the pipeline is IronClaw, which is focused on enhancing security protocols for AI agents and smart contracts alike.
Polosukhin is pushing toward creating what amounts to a secure AI operating system, one that inherently assumes adversarial conditions rather than bolting security on as an afterthought.
Decentralization as a security strategy #
Polosukhin consistently frames decentralization itself as a critical defense mechanism, particularly against the concentration of AI capabilities in the hands of a few major tech companies. If AI infrastructure is controlled by a handful of centralized entities, those entities become single points of failure. Distributing AI computation across decentralized networks makes the whole stack more resilient.
Polosukhin presented at Stanford’s Real World AI Security Conference in July 2026, focusing on vulnerabilities related to large language models and AI agents operating within blockchain environments.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our