How does CertAware use AI — and who’s in control?
AI helps CertAware keep training current with the scams happening right now — but a human approves every lesson before a member sees it, drafts are grounding-checked against their sources, no member personal information is sent to the model, and the AI never issues a verdict to a member. AI is the engine; people stay in control.
How the AI is governed
A human approves everything members see
AI drafts a lesson; a human reviewer approves it against its sources before it ever reaches a member. The accuracy gate never auto-approves — a person is always the final decision.
Grounding & faithfulness checks
Before a reviewer sees a draft, it is checked against its sources for faithfulness — including a deterministic detector that catches meaning reversals and negation flips, so a lesson can't silently contradict the source it came from.
No member PII goes to the model
The scam checker redacts obvious personal identifiers before anything is stored or sent to a model, and member personal information is never used to train models.
US-hosted, governed AI
Production AI runs on Amazon Bedrock in the US East (us-east-1) region — the same US footprint as the rest of the platform.
AI never issues a verdict to a member
The scam checker returns cautious warning signs and plain-English guidance, not a ruling. When a situation looks serious, it points the person to a human — it does not decide for them.
Auditable AI decisions
Approvals and rejections are recorded in the same tamper-evident audit trail as the rest of the platform, so you can show an examiner who approved which lesson, and when.
Built for vendor oversight
Examiners and your board increasingly ask how an AI vendor keeps its outputs accurate, private, and accountable. CertAware answers that with controls you can point to: a documented human approval step, grounding checks, PII minimization, US hosting, and an audit trail that records every approval decision. AI velocity is the reason your library stays current — human oversight is the reason you can defend it.
Reviewing our data handling? See our security page.