VIBE AI RED TEAMING
We red teamed a support agent. Guarded or not, it was breached.
A European cloud provider’s support agent was tested unguarded and with open-source guardrails. Reddy exposed data disclosure, unauthorized account actions, persistent prompt injection and internal-tool abuse in both.
European cloud provider · Support agentRead case study ↗
Unguarded configuration
15
Reported policy breachesAcross 13 attacks
Guarded configuration (open-source guardrails)
11
Reported policy breachesAcross 20 attacks
VIBE AI RED TEAMING
We red teamed a voice-activated kiosk. It turned into a safety issue.
A food-service chain’s ordering kiosk, built for blind customers, was tested before production. Reddy found that declared allergy and dietary restrictions were recorded for recommendations but not enforced when items were added to the cart.
Food-service chain · Voice-activated kioskRead case study ↗
Testing scope
16
Scenarios runAcross 4 topics
Result
6
Failures foundOut of 16 scenarios