AI red teaming — topics
LLM- and agent-shaped attack surfaces. See ai-red-teaming for path ordering.
Foundations
Prompt-layer
- direct-prompt-injection · indirect-prompt-injection
- jailbreaks · adversarial-suffixes
- multimodal-attacks
Agent and tool
- tool-confusion · mcp-attacks
- mcp-tool-poisoning-rug-pull
- agentic-tool-chain-confused-deputy
- copilot-zero-click-echoleak
- chain-of-trust-confusion
- exfiltration-via-rendered-content
RAG and memory
- rag-poisoning · memory-poisoning
- poisonedrag-five-doc
- phantom-rag-backdoor
- embedding-inversion-algen
Model-level
- model-extraction · training-data-extraction
- membership-inference
- supply-chain-attacks-on-models
- slopsquatting