How ruthless is AI as a Hacker? A $4.6 million vulnerability has been replicated, and two zero-day vulnerabilities have been discovered.

OnChainDetective

2025-12-02 03:22:23

Abstract generation in progress

[Block Rhythm] Anthropic recently pulled a tough stunt – letting AI act as a hacker, and the results are a bit scary.

They tested a batch of smart contracts that were actually hacked between 2020 and 2025 using Claude Opus 4.5, Sonnet 4.5, and GPT-5. These AIs directly replicated a vulnerability attack worth $4.6 million. Even more outrageous, after scanning 2,849 “seemingly fine” contracts, the AI actually discovered 2 brand new zero-day vulnerabilities and successfully simulated the profit process.

Data becomes more magical: The on-chain attack profits of AI have roughly doubled every 1.3 months over the past year. The research team's conclusion is very straightforward - these models are technically capable of autonomously seeking vulnerabilities and making money.

The security issues of smart contracts are already troublesome enough, and now with AI being misused, the risks of on-chain assets may need to be reassessed.

View Original

This page may contain third-party content, which is provided for information purposes only (not representations/warranties) and should not be considered as an endorsement of its views by Gate, nor as financial or professional advice. See Disclaimer for details.

12 Likes