Guardian watches a person's inbox, calls, voicemails, texts, card charges, receipts and breach feeds. Rules score each item and explain why; borderline cases get a model second look that can only raise concern, never clear it. Each action passes safety gates, then Guardian acts or waits:- Scam calls and texts (spoofed caller ID, gift cards, 'safe account', 'it's me, I'm in trouble'): alert first, then block, and tell a trusted contact who agreed to be told.
- Card misuse: lock only after a Persona live-human check; disputes always wait for the person.
- Phishing: quarantine. Account takeover and data leaks: exact steps.
- Live calls: a warning lands seconds before the money ask, then Block or 'I know this caller'.
Threats are ranked critical to low, nothing is ever repeated, and every step is on a readable audit trail.
The app: passkey sign-up, 'Just tell me' or 'Act for me', Needs you, 'This was real' (undo), push alerts with action buttons, read-only Gmail in watch-only mode, encryption at rest, Delete my data.
Sponsors: Amplitude (audit events), Statsig (7 safety gates + web analytics), SingleStore (threat store), Persona (human check), Superlinked (model second look), Firedrill (synthetic Gmail, Stripe, Slack).