Vitalik Buterin Warns: OpenClaw May Become an Entry Point for Data Leaks, Exposing AI Agent Security Risks

Gate News update: In 2026, Ethereum co-founder Vitalik Buterin issued a security warning about the popular AI development tool OpenClaw. He said that when it processes external data, it may have serious vulnerabilities, and users may experience data leaks or even have their systems remotely controlled without realizing it. As AI agent applications continue to roll out and accelerate in adoption, this issue has drawn strong attention from both developers and the security community.

According to the disclosed information, the core risk is that OpenClaw may execute hidden instructions when it reads webpage content. Attackers could craft malicious pages to prompt an AI agent to automatically download and run scripts, thereby stealing local data or tampering with system settings. In some cases, the tool quietly transmits sensitive information to external servers via commands like “curl,” and the entire process lacks warning prompts and auditing mechanisms.

Further security research suggests that this ecosystem risk has a certain degree of universality. Testing found that about 15% of “skills” (similar to plugin modules) contain potentially malicious logic. This means that even if the source appears trustworthy, it can still become an attack entry point. As developers quickly share functional modules, the lag in security review becomes more pronounced. When users install multiple skills on top of each other, the attack surface expands significantly.

Vitalik Buterin also emphasized that this is not a problem with a single tool, but rather a structural vulnerability widely present across the AI industry—feature iteration speed far outpaces the ability of security governance. He recommended reducing the risk of data exfiltration and systems being controlled by running models locally, isolating permissions, executing in sandboxes, and implementing approval mechanisms for critical actions.

Against the backdrop of AI agents gradually moving into software development and everyday scenarios, security has become a core variable. For users, they should avoid using plugins with unclear origins and strictly review permission requests. For developers, building a more comprehensive security framework will become part of long-term competitiveness.