NEW: ScamAvert for Android is here — get it on Google Play →
Live now

Meet SAM: ScamAvert’s new Scam Analysis Model

SAM looks at the wider situation, recognises the techniques scammers use and gives you a clearer warning when something may be wrong.

Already using ScamAvert? SAM has been released automatically, so there is nothing extra to install.

How SAM sees the situation

What it says
“Your device is infected”
Where it appears
A pop-up on an unfamiliar website
What it asks you to do
Call a support number immediately

SAM warning

Possible technical support scam

Genuine security warnings do not normally ask you to call an unknown support number.

What SAM is

A detection engine that reads the situation, not just the words

SAM stands for Scam Analysis Model. It is ScamAvert’s new AI scam detection engine, working alongside our other protection layers to understand suspicious activity in context and recognise the techniques scammers use.

  1. Signals

    What is on screen

    The text you can see, the request being made and any warning signs the app has already noticed.

  2. Context

    Where it is happening

    Which app or site the content appears in, and whether that request makes sense there.

  3. Scam technique

    What it resembles

    Whether the pattern matches a known technique: impersonation, pressure, a push to another channel.

  4. Clear warning

    What you should know

    A short, plain-language explanation of what looked wrong. Or no interruption at all when it does not add up to a scam.

Why we rebuilt it

Scams got more personal. Detection had to get smarter.

Today’s scams are patient, personal and good at looking legitimate: a refund notice that mentions a real retailer, a “security alert” in your bank’s colours, a delivery message on the day you are expecting a parcel.

Catching those means understanding what is happening, not only recognising familiar phrases. So instead of extending the previous engine, we built a new one around contextual scam detection.

Scam techniques SAM is designed to recognise

Examples of common techniques, not a complete list.

Refunds you never requested

“We have issued a refund for your order.” You did not place one, and the link wants your card details.

Fake device warnings

Pop-ups and ads claiming your computer is infected, “at risk” or needs urgent repair.

Impersonated support

Phone numbers, chats and “agents” pretending to be companies you already trust.

The push to another channel

“Let’s continue on WhatsApp.” Moving the conversation somewhere with fewer protections.

How SAM works

How SAM improves scam detection

Four differences you will notice in everyday use, without needing to know anything about what happens underneath.

Understands context, not isolated words

SAM considers which app or site the content appears in and whether the request makes sense there. Urgent wording on your bank’s own site reads differently from the same words in an unexpected email.

Checks suspicious situations more carefully

Content that raises concern goes through further verification before a warning is shown, which is designed to reduce false alerts on ordinary pages.

Distinguishes scams from discussions about scams

A news article about fraud, a forum post quoting a scam text or a block page from your own security software is not treated as an attack.

Explains warnings in plain language

Each warning includes a short reason: what looked wrong and what to be careful of. No threat scores and no jargon.

Built with users

Built using real customer feedback

SAM was shaped by what ScamAvert users told us: the scams that were missed, the warnings that were wrong and the messages they forwarded to us. Before switching it on for everyone, we tested and refined it with a group of those users first.

Alongside user testing, we compared SAM and the previous engine using the same curated internal benchmark of labelled safe and scam captures. SAM produced substantially fewer unnecessary warnings while continuing to identify the great majority of the defined scam cases.

This was an internal development benchmark rather than independent real-world certification. Results will vary depending on the content and situation being analysed.

How we tested SAM

Before release, we ran SAM and the previous engine against the same 1,374 labelled screen captures from our curated internal test corpus: 1,194 expected-safe captures and 180 expected-scam captures.

SAM reduced unnecessary warnings on the safe captures from 278 to 49. It identified 170 of the 180 defined scam captures, compared with 175 for the previous engine.

This is an internal development benchmark rather than independent real-world certification. Individual results will vary, and improving the remaining missed cases remains a priority.

  1. Reported

    Users sent us real scams, missed detections and warnings that should not have appeared.

  2. Tested and refined

    A group of ScamAvert users ran the new engine first, and their feedback shaped the final version.

  3. Released

    SAM is now active for everyone, with nothing extra to install.

Thank you to everyone who reported an issue or helped us test. You played a real part in making ScamAvert smarter.

Privacy

Smarter detection, the same privacy commitments

A better engine does not mean collecting more about you. Here is how ScamAvert handles what it analyses.

The full detail is in the ScamAvert privacy policy.

  • Checks start on your device. The ScamAvert apps run local checks first, and SAM is one layer of protection alongside them.
  • Text, not images. SAM analyses the text of what is on screen, cleansed of personal details before it is sent. The apps do not upload pictures of your screen.
  • Encrypted and processed securely. Text sent for analysis is encrypted in transit and processed on ScamAvert’s secure servers. Content from trusted sites and apps is cleared without AI analysis.
  • Used only to protect you. Analysis is used for protection and alerts. We never sell your data.
  • Your reports improve future updates. Reports from ScamAvert users help our team identify new scam techniques and improve future detection updates.

Availability

Where SAM is active today

SAM runs on ScamAvert’s servers, so the current versions of these products already use it.

ScamAvert for Windows

Emails, chats, pop-ups and web pages on your desktop.

SAM active

ScamAvert for Android

Messages, links and screens on your phone. About the Android app.

SAM active

Live Call Protection

Spots pressure and impersonation while a suspicious call is still happening. How Live Call Protection works.

SAM active

SAM is live. Put it to work.

Get ScamAvert’s protection for free, or choose Beta Plus to protect yourself while helping us keep ScamAvert available to others.

Found something we should investigate?

Report it to us. Your feedback helps our team improve ScamAvert’s future detection updates.

Report a scam or mistake
Common Questions

Frequently Asked Questions

Quick answers about SAM and what it means for you.

SAM stands for Scam Analysis Model. It is ScamAvert’s AI scam detection engine, designed to analyse suspicious content in context, recognise the techniques scammers use and provide clear, plain-language warnings when something may be wrong.
No. SAM runs on ScamAvert’s servers and has been released automatically, so current versions of the ScamAvert apps already use it. There is nothing extra to install.
The ScamAvert Windows desktop app, the ScamAvert Android app and Live Call Protection all use SAM.
SAM analyses the text of what is on screen, not images of it. The ScamAvert apps run local checks on your device, and the text they send for analysis is cleansed of personal details, encrypted in transit and processed securely on ScamAvert’s servers. Content from trusted sites and apps is cleared without AI analysis. Analysis is used only to protect you and send alerts, and it is never sold.
No. No security system can detect every scam, and SAM is designed to reduce risk rather than remove it. If something feels suspicious, pause, check independently and do not act under pressure, even if ScamAvert has not shown a warning.