OpenClaw's AI agents spark security fears with risky automation powers

A new AI tool called OpenClaw has quickly gained attention for its powerful but risky capabilities. Created by developer Peter Steinberger, the system uses agent-based automation to perform complex tasks. However, its broad permissions and ability to install new skills have raised serious security concerns.

OpenClaw operates as an AI agent with a distinctive crab logo. Users must grant it extensive access, including control over files, command execution, and web service accounts. The system can also 'wake up' and expand its functions by downloading additional skills—sets of instructions and code for new challenges.

Its rapid growth has led to an entire ecosystem around it. Websites now host skill downloads, while an agent social network, backup services, and hidden messaging channels have emerged. Yet, this flexibility has also made it a target for abuse.

Cybercriminals have exploited OpenClaw's skill-installation feature to distribute malware and steal cryptocurrency. The agents, powered by large language models, can identify vulnerabilities and automate attacks at scale. These models, while useful for defenders in analysing threats and generating reports, remain exposed to risks like prompt injection.

Despite these dangers, no documented changes to GitHub's security policies have appeared since OpenClaw's launch in early 2026. The tool's ability to run local shell commands and access sensitive data continues to worry experts.

The rise of AI agents like OpenClaw highlights the need for stronger cybersecurity measures. Security teams must develop advanced detection and response tools to counter automated threats. Without action, the risks of data breaches and financial theft will keep growing.