Key terms and concepts in AI agent security, threat detection, and observability. 25 terms and growing.
This glossary provides comprehensive definitions for key concepts in AI agent security. Whether you're building autonomous agents, securing LLM applications, or understanding the threat landscape, these definitions will help you navigate the field.
AI agent security encompasses the practices, tools, and technologies used to protect autonomous AI systems from attacks, prevent misuse, and ensure they operate safely within intended parameters. It addresses unique threats that emerge when AI systems can take real-world actions.
AI safety is the field focused on ensuring AI systems behave as intended and don't cause unintended harm. For AI agents, safety encompasses preventing dangerous outputs, maintaining alignment with user intentions, and ensuring agents can be controlled and corrected.
Anomaly detection identifies unusual patterns in AI agent behavior that deviate from established baselines. It helps catch novel attacks, unexpected behaviors, and system issues that signature-based detection would miss.
Autonomous agents are AI systems that can independently perform tasks, make decisions, and take actions without continuous human oversight. Their capability to act in the world creates unique security challenges around control, safety, and accountability.
Data exfiltration in AI agent security refers to the unauthorized extraction of sensitive information from an AI system or its environment. This can occur through prompt injection attacks, malicious tool usage, or compromised agents sending data to external endpoints.
Data privacy in AI systems concerns protecting personal and sensitive information that AI agents access, process, or generate. It encompasses data minimization, anonymization, access controls, and compliance with privacy regulations.
Defense in depth is a security strategy that layers multiple protective controls, so if one fails, others still provide protection. For AI agents, this means combining input validation, behavioral monitoring, sandboxing, and other controls to create comprehensive security.
Domain reputation is a scoring system that evaluates the trustworthiness of internet domains based on their history, ownership, content, and associations. For AI agents, domain reputation helps prevent interactions with malicious or suspicious websites.
Permission management controls what actions AI agents are allowed to perform, implementing least-privilege principles. It ensures agents have only the capabilities needed for their tasks, limiting potential damage from compromise or misuse.
PII (Personally Identifiable Information) detection identifies sensitive personal data in AI inputs and outputs. It helps prevent accidental exposure of names, addresses, phone numbers, social security numbers, and other identifying information.
Prompt injection is a security vulnerability where malicious instructions are inserted into AI prompts, causing the model to ignore its original instructions and follow the attacker's commands instead. It's one of the most critical security risks for AI agents operating autonomously.
Sandboxing for AI agents confines agent operations within isolated environments that limit access to sensitive resources. If an agent is compromised, the sandbox prevents it from affecting systems outside its contained environment.
Skill reputation is a trust scoring system for AI agent capabilities and plugins. It evaluates the safety and reliability of skills based on their source, behavior history, security audits, and community feedback.
AI telemetry is the automated collection and transmission of operational data from AI agents, including performance metrics, behavioral data, errors, and security events. It enables monitoring, debugging, and security analysis of agent systems.
Threat detection in AI agent security involves identifying malicious activities, attacks, or anomalous behaviors in real-time. This includes detecting prompt injection attempts, data exfiltration, unauthorized actions, and behavioral anomalies that could indicate a compromised or misbehaving agent.
Threat intelligence for AI agents is information about current and emerging threats, attack techniques, and malicious infrastructure. It helps security systems stay updated on the latest risks and enables proactive defense against known attacks.
Tool monitoring tracks and controls how AI agents use their available tools and capabilities—such as web browsers, code execution, file access, or API calls. It ensures agents use tools appropriately and detects suspicious or unauthorized tool usage.
Moltwire provides real-time security monitoring and threat detection for AI agents in production.
Maintained by the Moltwire Team
Last updated: February 2026 · 25 terms