#BOT #GPT #AI模型

What is a pleasant surprise? This is it:

A user on x stated that while using ChatGPT to generate code for developing a blockchain automated trading bot, the code recommended by GPT contained a backdoor that sent the private key to a phishing site, resulting in a loss of approximately $2500.

Yuxian, the founder of Slow Fog, confirmed in a post that there are indeed cases where AI-generated code has been hacked. Such attacks may stem from malicious patterns learned by AI from phishing posts or unsafe content, and current AI models find it difficult to discern whether there are backdoors in the code, necessitating caution against blindly trusting AI-generated code.