11 — Technology category · TR-110.80
AI Security & Guardrails
Security tools for AI applications and agents, at run time and before release, that inspect prompts and responses for injection and leaks, block tool calls a manipulated prompt or poisoned tool triggers, attack-test models and agents, and scan model files and AI components.
- AI Firewall
- LLM Guardrails
- AI Attack Testing
In scope
- Prompt and response inspection for injection and data leakage
- Blocking tool and MCP calls that a prompt or poisoned tool has manipulated
- Automated attack testing of models and agents
- Scanning of model files and AI components
Out of scope
- Quality and accuracy scoring of output over time (see TR-110.60)
- Staff use of public AI tools (see TR-100.80)
- Routing, keys, quotas and tool entitlements for model calls (see TR-60.60)
- AI inventories, risk assessments and regulatory mapping (see TR-110.70)
- General application and code scanning (see TR-100.50)
Example technologies · 5
Well-known technologies in this category, named as facts. Not a shortlist or an endorsement.
- Cisco AI Defense
- Palo Alto Networks Prisma AIRS
- HiddenLayer
- Lakera Guard
- NVIDIA NeMo Guardrails