OpenClaw AI agents cause data loss and operational disruption
OpenClaw, an open-source AI Agent platform deployed by major Chinese tech companies, reportedly experienced service crashes, memory loss, and operational failures. The platform's deep system-level permissions exposed users to security vulnerabilities including privilege bypass and remote code execution, with risks of data exposure and unauthorized device control.
What happened
OpenClaw, an open-source AI Agent platform, was reported to encounter service crashes, memory loss, and failure to complete tasks unless forcibly stopped. Users noted instances where OpenClaw did not finish tasks without manual intervention.
The platform was exposed to security vulnerabilities including privilege bypass and remote code execution flaws. These vulnerabilities could allow attackers to control user devices through malicious skill packages.
High-risk permissions granted to OpenClaw include the ability to read and write local files, operate browsers, and execute terminal commands. Users' large language model API keys, account information, and sensitive data were at risk of exposure if management was improper.
Reported impact
Users experienced service instability, with some reporting that 'Wukong', an Alibaba implementation, frequently failed to complete tasks and required forced stopping. Memory loss problems were documented alongside higher-than-expected computational resource consumption.
Medium-scale tasks such as calendar organization and email replies consumed five thousand to ten thousand tokens per task. Long-term memory functionality alone could consume over one hundred thousand tokens daily, resulting in daily baseline costs exceeding one hundred yuan at Chinese model pricing standards.
Index analysis
OpenClaw's core advantage—local deployment and data privatization—created inherent security risks. Granting AI systems deep permissions equivalent to 'system keys' exposed users to potential unauthorized access and data breaches if vulnerabilities were exploited.
The platform's autonomous decision-making and execution capabilities, while enabling independent task completion, increased risks of unintended actions affecting user systems and data. The combination of broad permissions and autonomous operation created attack surfaces for malicious actors.
Governance controls
Companies such as Tencent's WeChat integrated OpenClaw with stated capabilities to intercept and block risky situations, retaining the right to enforce rules. Future recommendations included implementing local context with cloud-side monitoring, constructing API key protection systems, and optimizing permission hierarchies with default closure of high-risk permissions.
Guidance for users recommended installing on non-primary machines to isolate core data, encrypting important and sensitive data, setting access permissions, and refraining from inputting bank card information, passwords, or trade secrets into OpenClaw. Regular security checks and avoidance of skill packages from unknown sources were advised.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.