NVIDIA releases tool to protect AI agents

VirusTotal research has revealed serious vulnerabilities in the artificial intelligence ecosystem. An analysis of 3,600 skills showed that one in five contains malicious code.

Attackers use hidden characters to mask commands. This allows them to stealthily steal API keys and forward personal data to third-party servers.

The detected threats include running malicious scripts from suspicious websites. Developers note the difficulty of detecting such attacks using standard code verification methods.

To address this issue, NVIDIA released the SkillSpector tool. This free, open-source solution is designed to scan skills before installation.

The service evaluates risk levels on a scale from 0 to 100 points. This system enables users to make informed decisions about the security of AI tools they use.

Developers are advised to regularly check their projects for vulnerabilities. Using specialized scanners helps minimize the risks of cyberattacks.