Threat Landscape & Frameworks

Overview

AI threats don't fit neatly into traditional cybersecurity taxonomies. They span the entire ML pipeline — from training data to inference output — and require frameworks designed specifically for machine learning systems.

Threat Actor Profiles

ActorMotivationTypical AttacksResources
Script kiddieCuriosity, bragging rightsKnown jailbreaks, copy-paste injectionLow — public tools only
Red teamerAuthorized testingFull methodology, custom toolingMedium-High — scoped access
CybercriminalFinancial gainAI-powered phishing, deepfakes, fraudMedium — cloud compute, social engineering
CompetitorIP theftModel extraction, training data theftHigh — funded research teams
Nation-stateEspionage, disruptionData poisoning, supply chain, influence opsVery High — custom labs, insider access
InsiderVariesTraining data manipulation, model backdoorsHigh — direct pipeline access

Key Frameworks

Two frameworks matter most for AI red teaming:

OWASP LLM Top 10

Focuses on application-level vulnerabilities in LLM deployments. Best for scoping pentests and communicating risk to developers.

OWASP LLM Top 10 Deep Dive

MITRE ATLAS

Focuses on adversarial tactics and techniques across the ML lifecycle. ATT&CK-style matrix for machine learning. Best for threat modeling and mapping attack paths.

MITRE ATLAS Deep Dive

Mapping to the Kill Chain

Cyber Kill Chain PhaseAI-Specific Activity
ReconnaissanceFingerprint model, extract system prompt, enumerate tools
WeaponizationCraft adversarial prompts, build injection payloads, fine-tune attack model
DeliveryPlant indirect injection in documents, web pages, emails
ExploitationExecute prompt injection, jailbreak, trigger backdoor
InstallationAchieve persistence via poisoned RAG source, tool manipulation
Command & ControlExfiltrate data via tool calls, establish ongoing injection channel
Actions on ObjectivesData theft, unauthorized actions, model compromise, disinformation