AgentGuard Research
Practical guidance, independent reviews, and clear explanations for teams building AI agents that can take real action.
ISO/IEC 42001 is an international standard for establishing, implementing, maintaining, and continually improving an artificial intelligence management system (AIMS). It sets management-system requirements for organizations that develop, provide, or use AI. It does not prescribe a particular model architecture or guarantee that an AI system is secure.
Read articleThe NIST AI Risk Management Framework (AI RMF) is voluntary guidance for managing risks associated with AI systems. It helps organizations organize decisions about trustworthy AI across design, development, deployment, and operation; it is not a product certification or a security control catalogue.
Read articleMITRE ATLAS is a knowledge base of adversary tactics and techniques for attacks on AI-enabled systems. It gives teams a shared way to describe how an attacker may influence models, data, training pipelines, retrieval systems, or deployed AI applications.
Read articleAn AI agent harness is the software layer that turns a model into an operating agent. It assembles instructions, context, tools, memory, execution rules, evaluation hooks, and approvals, then decides how a model response becomes the next step in a workflow.
Read articleAn agentic browser is a browser environment that lets an AI agent observe web pages and take actions such as navigating, filling forms, clicking controls, downloading files, or extracting results. The defining feature is an action loop: page state informs the next action, and the action changes what the agent sees next.
Read articleA rogue agent is an AI agent operating outside the organization’s approved ownership, authorization, or governance boundary. The label describes an unmanaged capability. It does not mean that the model is malicious, that an attacker has taken control, or that every unexpected response is an attack.
Read articleAn AI bill of materials (AI-BOM) is a record of the components and dependencies that shape an AI system: models, datasets, prompts, software packages, tools, services, and deployment configuration. Its job is traceability. It does not assess every component for safety or prevent an agent from taking an action.
Read articleDefine AI jailbreak, its boundary from prompt injection and red teaming, representative mechanisms, risks, controls, and safe evaluation.
Read articleDefine an MCP server, its primitives, request flow, deployment forms, boundaries, examples, and security considerations without becoming an implementation guide.
Read articleUnderstand shadow AI, how it differs from shadow IT, where it appears, which risks it creates, and how to discover and govern it.
Read articleLearn how MCP hosts, clients, servers, capabilities, transports, and downstream systems work, including the security responsibilities MCP leaves open.
Read articlePrompt injection lets untrusted text override an AI system’s intended instructions. It may come directly from users or indirectly through webpages, documents, emails, and tools. Defenses should separate data from authority, restrict permissions, validate actions externally, require approval for sensitive operations, and log decisions. Simple phrase filtering is insufficient protection.
Read article