← All practice areasProjects · 3 projects

AI assurance test beds

Systems I built myself to test the controls I recommend to clients: an internal audit agent, an AI security agent and the live demo for the Singapore AI governance case study. In each one the model reads and drafts, code does the numbers, and a named person decides. None of them is a client deployment.

Still life of a brass microphone, a navy notebook, a security key and a sketched three-step flow on a linen table

AI assurance test beds projects

Still life of a brass microphone, a navy card wallet and a security key
AI assurance test bed built by me

Bank of the Future: Voice Banking Agent

Working system Client: Confidential

Customers pay bills, add payees and dispute charges by voice, while 14 written policy rules decide what the agent may do and hold likely scams.

318Test scenarios run
0Wrong actions in evaluation
0 of 30Prompt injections succeeded
See how it works
Gold light on a dark horizon over a city waterfront at night
AI assurance test bed built by me

AI Security Voice Agent

MVP in assessment Client: Confidential

Security officers on duty get the right procedure by voice, drawn only from their own site's approved instructions, and every report is approved before it is sent.

163 of 164Reference test cases passed
68 of 69Independent blind cases passed
1.9 sMedian response on the hosted system
See how it works
Still life of an audit binder, a loupe and a fountain pen on a desk at dusk
AI assurance test bed built by me

Agentic Internal Audit System

Working system Client: Confidential

Audit teams cover ISO 27001, ISO 42001 and SOC 2 readiness with AI doing the reading and drafting, and an auditor grading every finding.

10 of 10Hidden instructions caught after fixes (7 on the first run)
6 of 6Evaluation suites passing
EveryFinding graded by a named auditor
See how it works
Other practice areas