Beta · Concierge Pre-empt Audit

We do what the scammer would do — before they can — so you can shut the door.

Vigil is a household digital-security service. A person runs the exact outside-in attack an AI scammer would run against your own surface — using this week’s real breach data — then hands you a plain list of the open doors to close.

Your own surface only. Run with your signed consent. No one else’s systems are ever touched.

Consent-first, default deny Plain-language shut list This week’s real breach data
Why now

The attackers already automated. The doors are already open.

These are real reports from the last week of November 2026. We cite them as reported — scope that a company has not confirmed is marked as unconfirmed.

Session tokens exposed

Wyndham: login tokens left reachable

A password hardcoded in Wyndham’s Android app reportedly opened an internal database holding member names, emails and JWT login tokens — the kind that keep an account signed in. Wyndham has roughly 129M enrolled rewards members; the number actually exposed is unconfirmed.

Reported by SC World, Nov 2026

SMS 2FA broken

125M+ text messages leaked — with the codes

Researchers found an exposed database at CSC Telecom (Baltics) holding 125M+ SMS messages, including two-factor codes and password-reset links sent on behalf of named banks. If a code lands in the leak, SMS 2FA stops being a wall.

Reported by Cybernews, Nov 2026

AI runs the attack

An AI agent broke into ~10 organizations

Palo Alto Networks (Unit 42) tracked a campaign that used an AI agent to run the intrusions largely on its own — reaching unauthorized access at around 10 organizations and pulling more than a million records at one. The attacker is now a program that works at machine speed.

Reported by CSO Online, Nov 2026

Voice-clone fraud

$129M stolen with cloned executive voices

U.S. prosecutors charged three men who allegedly used AI to clone the voices — and sometimes the image — of executives and lawyers, convincing companies to wire about $129M. The “person” on the call was synthetic.

Reported by ITV News, Nov 6 2026

Baseline context: the FBI’s IC3 logged 5,100+ bank-impersonation account-takeover complaints and $262M+ in losses heading into the holidays. (FBI / IC3, Nov 2025)

How the audit works

Three steps. One operator. Your surface only.

This is a rep-led beta: a real person onboards your household and runs the audit with you.

1

You sign the consent

We scope the audit to your household’s own surface and nothing else. You sign a plain consent that says exactly what we will and will not do. You can withdraw it at any time.

2

We run the attack & the checks

An operator runs the same outside-in moves a scammer would — matched against this week’s real breach data — to find which of your doors are actually open. We look; we never walk through.

3

You get the shut list

You receive a plain-language list of the open doors, ranked by what a scammer would reach for first, with the specific step to shut each one. No jargon, no dashboard to babysit.

What it never does

The limits are the product.

A security audit you can’t fully trust is worse than none. Here is what Vigil will never do.

Your own surface only

We test what belongs to your household. We never touch anyone else’s accounts, devices or systems.

Never signs in as you

We find the open door and stop. We don’t log into your accounts or read your mail to prove a point.

Never moves money

No transfers, no purchases, no account changes. We report the risk; you decide what to do.

Nothing without consent

Every action is checked against your signed consent first. Default is deny. Withdraw it and we stop.

Beta · limited households

Request your audit (beta)

Tell us how to reach you. An operator replies to set up your consent and schedule the audit. This is an early proof of concept, run by hand on a small number of households.

Opens your email app, addressed to matt@happyrobots.com, with your details filled in. We only use it to arrange your audit.

Prefer to write us directly? matt@happyrobots.com