How 1.5M Credentials Were Stolen
In February 2026, the Moltbook platform suffered a catastrophic security breach. Attackers exploited a vulnerability in the Oracle skill to steal over 1.5 million credentials and $400,000+ in API usage.
The attack was discussed extensively on x.com (Twitter), with the security community revealing how threat actors targeted AI agent platforms.
Before/After Case Study
BEFORE: Skill installs, gets full access to all environment variables. Attackers steal credentials undetected.
AFTER (ASF): Skills must request specific permissions. Credential access is logged and audited. Malicious access is blocked.
The Vulnerable Code
api_key = os.environ.get("OPENAI_API_KEY")This single line allowed any process to read ALL credentials stored in environment variables.
ASF Prevention
- Secure Credential Storage: ASF implements encrypted credential management with permission-controlled access
- Capability Enforcer: Prevents skills from accessing sensitive APIs without explicit authorization
- Skill Security Scanner: Automatically scans all skills before installation, flags environment variable access
- Zero-Trust Architecture: No implicit trust, least privilege, comprehensive audit logging
Reference: x.com security discussions on AI agent vulnerabilities (Feb 2026)