← All articles

Fake AI apps and extensions: checking a tool before you install it

Everyone wants to try the latest AI tool. Scammers know this. Which is why one of the traps that has spread fastest this year is not a fake email or a phone call: it is an app or a browser extension pretending to be a known AI tool — or promising "ChatGPT free, no limits" — that reads everything you do once installed.

You do not have to be a large company for this to happen. It is enough for someone on the team to search for a tool that summarises documents and click the first result.

A case that could be yours

An accountancy firm, six people. An administrator searches for "AI PDF summariser" to get through client contracts faster. The first result is an ad: a browser extension, good logo, thousands of reviews, free. She installs it in under a minute.

The extension does what it promises: it summarises PDFs. It also holds permission to "read and change all your data on the websites you visit". That includes the webmail, the cloud accounting software, and the government portal where the firm files on behalf of clients. For weeks, everything passing through the browser goes out to an unknown server.

Nobody notices until a client receives a scam email quoting, in precise detail, an invoice only the firm knew about.

Why it is easier to fall for now

Warning signs before you install

The five-minute checks

  1. Go to the official site yourself, typing the address, without going through a search engine or an ad, and install from there.
  2. Read the permissions before clicking. If you cannot see why it needs them, do not install it.
  3. Search the tool's name plus "scam" or "malware". If others have been hit, you will find it.
  4. Ask where the data goes. Any AI tool handling client documents should say plainly where it stores them, whether it uses them to train models, and whether the servers are inside the European Union. If it does not say, do not put anybody's data in it.

One simple rule for the whole organisation

The measure that protects most is not technical: agree that nobody installs AI tools on a work machine without a designated person having reviewed it. That is not distrust; it is the same control you already apply to who can make payments.

Pair it with a short list of approved tools, so the team has alternatives and does not go looking on its own. Banning everything without offering anything just moves the behaviour out of sight — and shadow IT you cannot see is worse than shadow IT you have a list of.

If you have already installed one

  1. Uninstall it from the browser or device immediately.
  2. Change the passwords of anything you used in that browser, starting with email, and close all open sessions.
  3. Check the mail rules and forwarding settings for anything you did not create.
  4. Tell the rest of the team, including which tool it was. The one that got installed once gets installed again two desks away.

This article is general information, not security advice for a specific case.

About this article

iaetica.cat is a space for writing in the open: what I am learning about artificial intelligence, its ethics and its law.

— Laura Graells