Definitive reference pages on agentic AI security: what each term means, the risks it creates, the architecture involved, the controls that work, and the sources behind every claim.
Agentic AI security protects autonomous AI agents, the tools they call and the AI Principals that create and command them. Definition, risks, architecture, controls and FAQ.
Read the explanationAI agent security protects individual AI agents — their identity, tools, memory and actions — from compromise and misuse. Definition, threats, architecture, controls and FAQ.
Read the explanationShadow AI is any AI tool, model or agent used inside an organization without security, privacy or governance approval. Definition, risks, examples, controls and FAQ.
Read the explanationInjected AI is agentic AI pushed into your environment by a trusted SaaS vendor — enabled by default, granted your existing permissions, and never reviewed. Risks, examples and controls.
Read the explanationAn AI Principal is code with creation authority: it spawns AI agents, trains their models and issues their commands. Definition, types, attack scenarios, controls and FAQ.
Read the explanationAn agentic SOC uses autonomous AI agents to triage, investigate and respond to security alerts alongside human analysts. Definition, architecture, benefits, risks and FAQ.
Read the explanationLLM security protects model inputs and outputs. AI agent security protects actions, identities, tools and memory. A side-by-side comparison with risks, controls and FAQ.
Read the explanationWhat an AI agent security platform must do: discovery, identity, tool control, runtime tracing, containment and creation-layer coverage. Categories, criteria and buyer FAQ.
Read the explanationThe main agentic AI security risks — prompt injection, excessive agency, shadow agents, memory poisoning, tool abuse, multi-agent cascades and Principal compromise — with mitigations.
Read the explanationHow OWASP's GenAI Security Project frames agentic AI threats and mitigations, how they relate to the OWASP LLM Top 10, and how to operationalize them in an enterprise.
Read the explanationModel Context Protocol connects AI agents to tools and data. MCP security covers server trust, tool-definition integrity, credential scope, consent and monitoring. Risks and controls.
Read the explanationAI agent governance defines who may create agents, what they may access, how they are reviewed and how they are retired. Framework, policies, controls, metrics and FAQ.
Read the explanation