SlowMist and Bitget Jointly Release AI Agent Security Report, Revealing Potential Risks of Web3 Automated Trading

As AI Agent applications in crypto trading heat up, automated trading is shifting from tool-assisted to autonomous execution, but security risks are emerging simultaneously. Recently, security firm SlowMist and exchange Bitget jointly released an AI Agent security report, systematically sorting out potential threats and protection systems for Agent automated trading in Web3 scenarios.

Combining real cases, the report analyzes typical security issues faced by AI Agents: behavior manipulation via prompt injection, supply chain risks in the plugin ecosystem, API Key abuse and permission expansion, and misoperations caused by automated execution.

The report recommends users control permissions (e.g., sub-account isolation, API IP whitelisting), establish transaction monitoring and anomaly alert mechanisms; high-risk operations require manual confirmation or independent signatures to avoid asset losses. A transaction security self-check list is attached at the end to help users quickly identify hidden risks.