When a hacker tries to trick one AI, they accidentally train a second AI to be a better bodyguard.
April 10, 2026
Original Paper
Pre-Execution AI Gating for Smart Contracts: Multi-Model Consensus with Adversarial Verification
SSRN · 6267278
The Takeaway
This 'inverse effect' discovery allows for a multi-model consensus system that secures smart contracts on the blockchain. When an adversarial attack succeeds in confusing one architecture, it often pushes another model into a state of higher clarity, creating a novel and robust defense mechanism.
From the abstract
AI agents are now making financial decisions on blockchains without human oversight. They move fast, executing transactions in milliseconds, but when they make mistakes, those mistakes are permanent. In 2025 alone, this resulted in over $3 billion in stolen funds. Worse, attackers can now scan and exploit smart contracts for just $1.22 each [SCONE-bench], and their capabilities are doubling every 1.3 months. Currently, nothing checks whether an AI's decision is correct or manipulated before it b