OpenClaw has partnered with Google-owned VirusTotal to scan skills uploaded to its ClawHub marketplace, strengthening defenses against malicious content in its agentic AI ecosystem.
Each skill is assigned a unique SHA-256 hash and checked against VirusTotal’s threat intelligence database. If no match is found, the code is uploaded for deeper analysis using VirusTotal’s Code Insight feature. Skills marked benign are automatically approved, suspicious ones are flagged with warnings, and malicious skills are blocked. OpenClaw also re-scans all active skills daily to detect newly emerging threats, though it cautions that the system is not foolproof, especially against advanced prompt injection attacks.
The move comes after researchers discovered hundreds of malicious skills disguised as legitimate tools. These skills were capable of stealing data, installing backdoors, and executing unauthorized commands. Security analysts warn that AI agents with system-level access can become covert data-exfiltration channels and execution orchestrators, bypassing traditional security tools.
OpenClaw’s growing popularity, along with its companion social platform Moltbook, has raised serious security concerns. Because skills can control devices, manage finances, and interact with external services, they significantly expand the attack surface and can turn the platform into an “agentic trojan horse” if abused.
Researchers have identified multiple vulnerabilities, including plaintext credential storage, prompt injection exploits, exposed APIs, misconfigured databases leaking millions of tokens, and remote code execution flaws.
Experts warn that agent platforms are spreading rapidly across enterprises without formal security oversight, creating a new category of Shadow AI risk. Regulators and cybersecurity firms are urging organizations to implement strict configuration controls, identity and access management, and sandboxing to prevent AI agents from becoming powerful attack vectors.
Comments