Malicious AI agent skills can be packaged to steal credentials, exfiltrate source code, and install backdoors while still bypassing many current skill-auditing systems. The paper finds that static scanners are especially weak against payload-preserving evasions, while runtime behavior auditing is far more resilient. The core threat is simple but serious: an agent skill is not […] The post Malicious Agent Skills Can Steal Credentials, Exfiltrate Source Code, and Install Backdoors appeared first on GBHackers Security | #1 Globally Trusted Cyber Security News Platform.
First seen on gbhackers.com
Jump to article: gbhackers.com/malicious-agent-skills/
