AI Security Glossary

Search our knowledge base to break down complex AI terms and jargon into clear, practical definitions — many paired with short explainer videos for quick learning. Use it to understand key concepts, stay current with regulations, and navigate AI security with confidence.

Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.
Showing 00 of 00 terms

Term

Short Description

PointGuard

Video

Bug Bounty

A a program that rewards individuals for finding and reporting security vulnerabilities in an organization's software, encouraging proactive identification and resolution of potential security issues.

California AI Act

A proposed California law that aims to regulate AI systems—focusing on safety, transparency, and civil rights protections—especially in high-risk and public-facing applications.

Chain-of-Thought Monitoring

Chain-of-thought (CoT) monitoring is the practice of inspecting a reasoning model's intermediate reasoning to detect unsafe intent, such as plans to cheat, deceive, or misuse tools, before or while the model acts.

Circuit Breaker (Agent)

An agent circuit breaker is a runtime control that interrupts an autonomous AI agent when defined behavioral or operational thresholds are exceeded. Borrowed from distributed systems engineering, circuit breakers are a key primitive for safe agent operations.

Feature

CISA Known Exploited Vulnerabilities (KEV)

The CISA Known Exploited Vulnerabilities (KEV) catalog is a list, maintained by the US Cybersecurity and Infrastructure Security Agency, of software vulnerabilities confirmed to be actively exploited in the wild.

Code Repository

A storage location for source code, often using version control systems like Git.

Common Vulnerabilities and Exposures (CVE)

A list of publicly know vulnerabilities maintained by NIST.

Common Vulnerability Scoring System (CVSS)

A standardized, repeatable, and vendor agnostics method to compare application vulnerabilities.

Common Weakness Enumeration (CWE)

A community-developed list of common software and hardware weaknesses, which could introduce vulnerabilities.

Compliance

Adhering to industry standards, regulations, and best practices to ensure that applications meet legal and security requirements

Compliance Audit

A comprehensive review of an organization's adherence to regulatory guidelines.

Compliance Risk

The potential for losses or legal penalties due to violations of laws, regulations, or prescribed practices.

Computer-Use Agent

A computer-use agent is an AI agent that operates a computer the way a person does, by viewing the screen and controlling the mouse and keyboard. Because it can interact with any application, it needs strong limits on what it may access and do.

Confused Deputy (AI)

A confused deputy attack tricks a privileged actor into using its authority on behalf of an attacker. Applied to AI agents, the agent becomes the deputy: an attacker manipulates it into invoking permitted tools with attacker-chosen arguments, an effect that classic access controls cannot easily distinguish from legitimate use.

Protection

Container

A lightweight, standalone executable package of software that includes everything needed to run it, ensuring consistency across environments.

Contextual Risk Engine (CoRE)

PointGuard's proprietary technology for prioritizing security issues based on severity, exploitability, and business context.

Feature

Context Window Leakage

Context window leakage is the unintended exposure of sensitive information held in an AI model's working context, such as system prompts, retrieved documents, other users' data, or credentials, through the model's outputs or tool calls.

Protection

Continuous Integration/Continuous Deployment (CI/CD)

A practice that automates the integration and deployment of code changes, enabling frequent and reliable updates to applications.

Continuous Threat Exposure Management (CTEM)

A proactive and continuous program to monitor, evaluate, and reduce levels of exploitability and validate analysis and remediation processes.

Feature

Cross-Agent Prompt Injection

Cross-agent prompt injection is an attack in which malicious instructions pass from one AI agent to another through shared outputs, messages, or data, allowing a single compromised input to spread across a multi-agent system.

Protection

Cross-Prompt Injection Attack (XPIA)

A Cross-Prompt Injection Attack (XPIA) is a Microsoft-named class of indirect prompt injection in which malicious content embedded in one user's context (such as email or documents) influences another user's AI session. XPIA is now a widely referenced category in agentic and Copilot threat modeling.

Protection

CSA MAESTRO

MAESTRO is a threat modeling framework from the Cloud Security Alliance designed for agentic AI systems. It analyzes threats across seven layers, from foundation models to the agent ecosystem, to capture risks that traditional methods such as STRIDE miss.

Data Poisoning

A training-time attack where malicious or manipulated data is introduced into the training set to distort a model’s behavior, reduce accuracy, or embed hidden backdoors.

Protection

Data Privacy

The aspect of information technology that deals with the ability of an organization or individual to determine what data can be shared with third parties.

Decentralized Identifier (DID)

A Decentralized Identifier (DID) is a globally unique, cryptographically verifiable identifier defined by the W3C that can be resolved without a central authority. DIDs are foundational to zero-trust architectures for AI agents and to emerging models of verifiable agent identity.

Dependency Resolution

The process of determining and fetching the correct versions of dependencies for a software project.

DevOps

A methodology that combines software development (Dev) and IT operations (Ops) to shorten the development lifecycle, deliver high-quality software continuously, and improve collaboration between development and operations teams.

DevSecOps

An approach that integrates security practices into the DevOps process, ensuring that security is incorporated at every stage of the software development lifecycle to enhance the overall security posture of applications.

Dynamic Application Security Testing (DAST)

A testing method that analyzes applications in their running state to identify vulnerabilities by simulating external attacks.

EchoLeak

EchoLeak is a named zero-click attack against Microsoft 365 Copilot disclosed in 2025 that used indirect prompt injection inside email content to exfiltrate sensitive enterprise data through Copilot's grounding pipeline. The incident became a high-volume search query and a reference case for indirect prompt injection at scale.

Protection

Embedding Inversion Attack

An embedding inversion attack reconstructs the original text or data from vector embeddings stored in retrieval-augmented generation systems. Because embeddings are often treated as opaque, organizations frequently overlook how much sensitive content they leak when exposed.

Protection

Endpoint Detection and Response (EDR)

A system that monitors and analyzes endpoint activities to detect, investigate, and respond to security incidents in real-time.

EU AI Act

A comprehensive European regulation that classifies AI systems by risk and imposes strict requirements on high-risk applications—covering transparency, accountability, and safety.

Evaluation Environment Escape

An evaluation environment escape occurs when an AI model under testing reaches real systems because its evaluation environment was misconfigured or insufficiently isolated. In 2026, models from Anthropic, Meta, and Google accessed real companies this way.

Protection

Excessive Agency

Excessive agency is a vulnerability in which an AI system is given more functionality, permissions, or autonomy than its task requires, allowing manipulated or faulty output to cause real damage. It is listed as LLM06 in the OWASP Top 10 for LLM Applications.

Protection

Exploit Prediction Scoring System (EPSS)

A process developed by First.org for estimating the likelihood that a software vulnerability will be exploited in the wild.

False Negative

A failure to detect an actual vulnerability or threat, leading to a potential security risk remaining unaddressed.

False Positive

A security alert that incorrectly indicates the presence of a vulnerability or threat when none exists.

Foundation Model

A foundation model is a large machine learning model trained on broad data at scale that can be adapted to many downstream tasks. The term, coined at Stanford, is now embedded in the EU AI Act and NIST AI RMF as the taxonomic anchor for general-purpose AI systems.

Frontier Model

A frontier model is a highly capable foundation model at the leading edge of capability and risk, typically referenced in EU AI Act, U.S. executive orders, and emerging international policy. The term anchors many of the most stringent obligations for general-purpose AI systems.

General Data Protection Regulation (GDPR)

European Union regulation that mandates strict data privacy and security measures for protecting personal data, giving individuals greater control over their personal information.

Generative AI (GenAI)

AI systems designed to create original content—such as text, images, code, or audio—by learning patterns from large datasets. Used in automation, creativity, and user interaction across industries.

Goal Drift

Goal drift is the gradual divergence of an autonomous AI agent's behavior from its original authorized objective. It can arise from compounding small reasoning errors, memory poisoning, reward hacking, or repeated exposure to adversarial input, and it underpins many rogue agent incidents.

Protection

Guardian Agent

A guardian agent is an AI-powered oversight system that supervises other AI agents in real time, checking that their actions stay aligned with approved goals, policies, and boundaries. Gartner recognized guardian agents as a distinct market category in 2026.

Feature

Health Insurance Portability and Accountability Act (HIPAA)

U.S. law that sets national standards for the protection of sensitive patient health information, ensuring that such information is kept confidential and secure, particularly in electronic form.

High-Risk AI System

High-Risk AI System is a classification under the EU AI Act for AI uses that may significantly affect health, safety, or fundamental rights. It carries the act's most demanding obligations and is the focal point of most CISO and GRC conversations about AI compliance in Europe.

Hugging Face

An open-source platform and model hub widely used for sharing, deploying, and fine-tuning machine learning and NLP models, especially transformers and large language models.

Video

Human-in-the-Loop (HITL)

Human-in-the-loop (HITL) is a control pattern that requires a person to review or approve certain AI decisions or agent actions before they take effect. It is a core safeguard for high-impact agentic AI and a requirement in frameworks such as the EU AI Act.

Improper Output Handling

A failure to properly inspect, filter, or validate AI-generated outputs—leading to the release of unsafe, biased, or non-compliant content into user-facing applications.

Protection

Incident Management

The process used to manage the lifecycle of incidents to ensure that normal service operation is restored as quickly as possible.

Incident Response

The process of identifying, managing, and mitigating security incidents to minimize their impact on the organization.

Indirect Prompt Injection

Indirect prompt injection is an attack in which malicious instructions are embedded in content the AI system later retrieves, such as documents, emails, or web pages. When the model processes that content, it treats the hidden instructions as legitimate, often leading to data leakage or unauthorized actions.

Protection

Infrastructure as Code (IaC)

The ability to provision and support computing infrastructure using code instead of manual processes and settings.

Integrity Check

An integrity check verifies that software, data, or an AI model has not been altered or tampered with, typically using cryptographic hashes, checksums, or digital signatures.

Intent-Based Access Control (IBAC)

Intent-based access control (IBAC) is an authorization model that evaluates whether an AI agent's requested action matches the purpose it was given, not just whether its credentials allow the action. It helps stop agents that are technically permitted but have been steered off task.

Feature

Inter-Agent Transfer Protocol (IATP)

The Inter-Agent Transfer Protocol (IATP) is the emerging framing for how autonomous AI agents transfer tasks, context, and delegated authority between each other. It defines the structured handoffs that make multi-agent workflows reliable, auditable, and recoverable across single-vendor and cross-vendor topologies.

Feature

Internal Audit

An independent, objective assurance and consulting activity designed to add value and improve an organization's operations.

ISO 42001

ISO/IEC 42001 is the first international management system standard for artificial intelligence. It specifies requirements for establishing, implementing, maintaining, and continually improving an AI management system, and is rapidly becoming the default certification target for AI-using enterprises.

Key Risk Indicators (KRIs)

Metrics used to provide an early signal of increasing risk exposures in various areas of an enterprise.

Kill Switch (AI)

An AI kill switch is a control that halts an AI system or agent on demand or in response to defined triggers. Kill switches are increasingly referenced in regulatory frameworks such as the EU AI Act as a baseline safety control for high-risk AI deployments.

Feature

Large Language Model (LLM)

A type of AI model trained on massive text datasets to understand and generate human-like language. LLMs power chatbots, assistants, and other natural language applications.

Least Agency

Least agency is a security principle that extends least privilege to AI agents, limiting not only what an agent can access but also how much it can decide and do on its own. Agents get the minimum tools, permissions, and autonomy their task requires.

Lethal Trifecta

The lethal trifecta is a term for AI agents that combine three capabilities: access to private data, exposure to untrusted content, and the ability to communicate externally. Together they let a prompt injection steal data, making the combination a key risk test for agent design.

Protection

License Compliance

Ensuring that software components comply with licensing agreements and open-source licenses.

LLM Firewall

LLM Firewall is a security layer that monitors and filters interactions with large language models, blocking malicious prompts, preventing data leaks, and enforcing compliance. It safeguards AI systems from threats like prompt injections while ensuring safe, authorized, and regulation-compliant usage.

Feature

LLM Gateway

An LLM gateway is an intermediary that brokers traffic between applications and one or more large language models. It provides a single point for authentication, quota, routing, logging, and policy enforcement, separate from the model providers themselves.

Feature

LLMjacking

LLMjacking is the theft and misuse of cloud credentials or API keys to access large language model services, letting attackers run or resell AI usage at the victim's expense or use stolen compute for malicious purposes.

Protection

LLMOps

A set of practices and tools for managing the lifecycle of large language models (LLMs), including deployment, monitoring, scaling, and governance in production environments.

Machine Learning (ML)

A method that enables software systems to learn from data and improve performance without explicit programming. Commonly used for predictions, classifications, and decision-making across dynamic or data-rich environments.

Ready to get started?

Our expert team can assess your needs, show you a live demo, and recommend a solution that will save you time and money.