Guide/Artificial intelligence
ARIA: Agentic AI Security
ARIA is the module for assessing the security of AI agents and agentic AI systems, based on the AIVSS (AI Vulnerability Scoring System) framework.
What it assesses
- 8 risk categories (A1-A8) specific to agentic systems (e.g. prompt injection, privilege escalation, exfiltration via tools, memory/context manipulation)
- 35 mitigation controls mapped to those categories
- AIVSS score (0-10) per system/agent assessed, with detail by risk category
Usage flow
- Register the AI system or agent to assess (name, description, use cases, tools/access it has)
- Answer the agentic security controls organised by the 8 categories A1-A8
- The system calculates the AIVSS score and identifies the missing controls
- See the assessment detail for the remediation plan
ARIA complements the EU AI Act (section 36): while the EU AI Act covers regulatory obligations, ARIA focuses on technical security vulnerabilities specific to AI agents.