Blockchain security firm warns of AI code poisoning risk after OpenAI’s ChatGPT recommends scam API
Ad Yu Xian, founder of Slowmist, warns of AI code poisoning, where harmful code is injected into AI training data, posing risks to users. A recent incident involved a crypto trader losing $2,500 after ChatGPT recommended a fraudulent Solana API for creating a bot. This led to the theft of the trader's private keys and assets within 30 minutes. Investigation revealed the API's domain was registered two months prior and consistently received stolen tokens, indicating premeditated fraud. While the incident highlights vulnerabilities in AI tools, there is no evidence that OpenAI intentionally integrated malicious data. The rise of scams exploiting AI training data raises concerns about trust in AI-driven tools and potential financial losses for users.
