Major Security Flaws Found in Popular AI Platforms
AI Platforms Expose Sensitive Data Through Critical Flaws
Security experts are sounding the alarm after discovering gaping vulnerabilities in OpenClaw (formerly Clawdbot) and Moltbook, popular platforms in the AI ecosystem. These weaknesses could give attackers nearly unfettered access to sensitive data and system controls.
Shockingly Low Security Scores
Developer Lucas Valbuena's tests using ZeroLeaks revealed OpenClaw scored a dismal 2 out of 100 on security assessments. The platform proved vulnerable to:
- 84% success rate for data extraction
- 91% success rate for prompt injection attacks
"These numbers should terrify anyone using these platforms," Valbuena noted. "It's like leaving your front door wide open with a sign saying 'Take what you want.'"
Database Left Completely Exposed
The situation appears even worse for Moltbook, where security researcher Jamieson O'Reilly found:
- Entire databases accessible on public networks
- Secret API keys available to anyone
- Potential for impersonating high-profile users like Andrej Karpathy
The vulnerabilities affect servers globally, with 954 exposed Clawdbot instances identified across China, the U.S., Germany and other countries.
Expert Recommendations
While perfect protection remains elusive, cybersecurity specialists advise:
- Never store sensitive data directly in configuration files
- Use environment variables for key management
- Implement Cloudflare Tunnel or zero-trust login systems
- Conduct regular security audits using tools like ZeroLeaks
The discoveries highlight growing pains in the rapidly expanding AI agent ecosystem, where security often plays catch-up with functionality.
