DEFINITION 01
AI Agent Permissions
AI agent permissions are the actions and data access granted to an AI driven workflow through its accounts, tools, and connected services.
The CyberGuard field guide / Topics
AI Security.
Understand the risks around AI models, prompts, agents, data, and the tools connected to them.
15 published definitionsPractical examples and source links
← Back to the dictionary
THE DEFINITIONS
Explore AI Security.
Open a term for a plain-language explanation, a technical explanation, and a practical example.
DEFINITION 02
AI Assisted Phishing
AI assisted phishing uses generative tools to draft, personalize, translate, or scale deceptive messages intended to steal information or prompt harmful action.
DEFINITION 03
AI Red Teaming
AI red teaming is an authorized, structured effort to find harmful or unsafe behavior in an AI system before or during deployment.
DEFINITION 04
Adversarial Machine Learning
Adversarial machine learning studies attacks that manipulate an AI system or its data to change behavior, reveal information, or degrade performance.
DEFINITION 05
Data Prompt Leakage
Data prompt leakage is unintended disclosure of sensitive material supplied to, retrieved by, or generated within an AI application prompt context.
DEFINITION 06
Deepfake
A deepfake is synthetic or altered audio, video, or imagery made to depict a person saying or doing something they did not actually say or do.
DEFINITION 07
Hallucination
An AI hallucination is a generated statement presented as if factual even though it is unsupported, incorrect, or invented.
DEFINITION 08
Insecure Output Handling
Insecure output handling occurs when an application treats AI generated text as trusted code, markup, instructions, or commands without appropriate validation.
DEFINITION 09
Large Language Model Security
Large language model security is the protection of LLM applications, their data, tools, models, and users against misuse and unintended behavior.
DEFINITION 10
Model Exfiltration
Model exfiltration is unauthorized removal of model files, weights, parameters, or enough model behavior to reproduce a valuable system.
DEFINITION 11
Model Poisoning
Model poisoning is deliberate interference with a model or its update process so the resulting system behaves incorrectly or contains a hidden trigger.
DEFINITION 12
Model Supply Chain
Model supply chain security concerns the origin and integrity of models, datasets, dependencies, tools, and build steps used in an AI system.
DEFINITION 13
Prompt Injection
Instructions embedded in untrusted content that try to redirect an AI system away from its intended task.
DEFINITION 14
Retrieval Augmented Generation Risk
Retrieval augmented generation risk is the possibility that retrieved documents or embeddings mislead an AI answer, cross access boundaries, or expose sensitive data.
DEFINITION 15
Training Data Poisoning
Training data poisoning is manipulation of examples used to train or fine-tune a model to bias, degrade, or redirect its future outputs.
KEEP EXPLORING
Follow another topic.
Build context across the connected parts of cybersecurity.
Browse all other topics
Network Security 22Malware 18Web Application Security 17Business Cybersecurity 15Incident Response 15Cloud Security 14Core Cybersecurity 6Cryptography 4Identity Access 4Phishing Social Engineering 4Vulnerability Management 4Fraud Online Safety 3Privacy Data Protection 2Security Technologies 2Threat Intelligence 2Endpoint Security 1Ethical Hacking 1Governance Risk Compliance 1Wireless Mobile Security 1