Red-team an AI system
Stress-test an existing AI initiative against 8 adversarial scenarios (prompt injection, data exfiltration, jailbreak, excessive agency, regulatory hallucination, context poisoning, supply-chain compromise, insider misuse).
How it works
- 1Pick an initiative from the list below.
- 2On its detail page, click 'Re-assess' with the red-team toggle enabled (requires an access code in gated demo mode).
- 3The ARA red-team agent runs 8 tests against the proposed controls and judges each Blocked / Partially Blocked / Vulnerable / Not Applicable.
- 4Overall resilience score is stored against the initiative and appears in the Decision Pack.
Pick an initiative
| Initiative | Band | Verdict | |
|---|---|---|---|
Autonomous personal-loan approval (Australia) SAMPLE-01 SAMPLE |
Critical | Escalate to risk committee | Open initiative |
AI-assisted AML transaction monitoring SAMPLE-02 SAMPLE |
High | Approve with conditions | Open initiative |
HR policy RAG chatbot (global workforce) SAMPLE-03 SAMPLE |
Moderate | Approve | Open initiative |
Autonomous SOC response agent SAMPLE-04 SAMPLE |
Critical | Do not approve | Open initiative |
Radiology AI triage assistant (hospital) SAMPLE-05 SAMPLE |
High | Approve with conditions | Open initiative |