OpenClaw AI Agent Flaws Could Enable Prompt Injection and Data Exfiltration

CNCERT warned that OpenClaw — a popular open-source autonomous AI agent — has weak default security, privileged endpoint access, and is vulnerable to indirect prompt-injection (IDPI/XPIA) attacks that can leak sensitive data (including via link-preview exfiltration). Researchers and vendors have demonstrated practical exploitation paths, and threat actors have published malicious GitHub installers distributing info-stealers (Atomic, Vidar) and GhostSocks proxy malware; CNCERT and others recommend hardening deployments, isolating services, restricting management ports, and obtaining skills only from trusted sources.