On-chain analysts discovered an interesting detail: after the attack on Balancer, the contract code deployed by the Hacker surprisingly still contained console.log debugging information.

This is not too common.

Experienced attackers usually clear all testing traces before formal operations. But this time it's different - log outputs like "Done with" appeared in the code, resembling template code copied and pasted directly from some AI model.

Could it be that LLM helped write this? This speculation is not without basis. Recently, there are more and more signs indicating that some attackers are beginning to rely on large language models to generate exploit code. It is convenient, but the code generated by AI often carries obvious "factory settings"—for example, those debugging statements that were forgotten to be deleted.

Smart contract security audits may have gained a new dimension: identifying the characteristics of AI code.