AI Safety
A simple response to AI-assisted phishing, cloned voices, fake video, urgent payment requests, and messages designed to make you act before you think.
Do not click, pay, share a code, install software, or continue the call just because the message feels urgent.
Leave the message or call. Use a trusted number, saved contact, official app, or website you open yourself.
Verify the person, request, amount, and action through that separate channel before doing anything.
The simple list
The scammer wants speed, secrecy, and a single channel. Your defense is time, separation, and independent confirmation.
Recognize the pattern
The core warning signs did not retire. They put on a cleaner shirt.
Watch for unexpected account problems, fake invoices, delivery failures, shared documents, password resets, job offers, or urgent executive requests. Hover or inspect carefully, but verify through the official service rather than the embedded link.
A voice may sound like a child, manager, officer, doctor, bank worker, or vendor. Hang up and call the person or organization through a known number. Ask a question the caller would not know from public information.
Do not rely on visual realism alone. Confirm through another live channel and look for the business process that should surround the request: normal approval, known account, documented change, and a second person.
Any surprise change to bank details, payroll, password reset, remote access, account recovery, gift cards, or urgent transfer deserves a separate confirmation step.
For families and teams
A good plan should still work while someone is frightened, tired, embarrassed, or in a noisy parking lot.
Information boundary
When you are unsure, move the information up a risk level until the organization approves it.
Public information, approved templates, generic examples, and content already intended for broad release. Normal review still applies.
Internal operations, draft plans, non-public meeting notes, vendor details, or limited personal information. Use only in an approved tool with a clear purpose and review.
Passwords, recovery codes, payment data, health information, protected client records, legal strategy, secrets, regulated data, or anything the organization has not approved for AI. Do not upload.
Something happened
People delay reporting when they fear punishment. That gives the problem more time to move.
The FTC, FBI, and CISA recommend pausing, independently verifying through known contact information, protecting account codes, and treating unusual urgency or payment methods as warning signs.
Next useful step
Download the two-page guide or open the fuller starter kit with information classes, review rules, incident notes, and a 90-day adoption path.
