MAESTRO
OWASP AIVSS Core Security Risks
Interactive Learning Module
🎓 Free for Everyone: Take all training courses and exams at no cost
🏆 Premium Members: Pass 7+ exams (70%+ score) to unlock your professional certificate
📝 Pass 7+ exams with 70%+ scores to unlock certificate (0/7 passed)
Agent's interaction with tools results in aberrant operational outcomes
Agentic AI Tool Misuse occurs when an agent's interaction with externalized functionalities, including tools, capabilities, or resources, results in aberrant or detrimental operational outcomes. This can stem from compromised integrity of toolchain integrations, deficiencies in the agent's inferential capabilities, malicious injection of tool specifications, or lack of cryptographic provenance.
Malicious Code Execution: LLM agent manipulated into executing arbitrary code via interpreter tool
Tool Squatting: Fraudulent storage service intercepts and exfiltrates sensitive data
DoS via Tool Loop: Infinite invocation of resource-intensive tools exhausting system resources
MCP Server Backdoor: Rogue Model Context Protocol server injecting persistent backdoors
Metadata Manipulation: Hidden instructions in tool metadata manipulating agent actions
Rogue Trading Agent: Financial agent executing unauthorized trades without effective kill switch