More and more people are letting AI do their shopping for them. Ask it to find the best price, fill the cart, sometimes even pay. Perplexity’s Comet browser does exactly that, and other tools are following the same path. The problem is this AI doesn’t suspect anything. It sees a page, it carries out the task, and it moves on to the next one.
A security lab, Guardio Labs, just proved it in black and white with a series of tests called Scamlexity. The results are embarrassing: the agent bought a watch from a fake site mimicking Walmart, entered banking credentials on a fake Wells Fargo login page, and downloaded a booby trapped file hidden inside a fake CAPTCHA. Every time, without asking a single question and without anyone noticing.
Quick answer: AI-powered browsers can get tricked just like a distracted person: buying from a fake store, typing a password into a phishing page, downloading a malicious file. For a small business in Quebec, that means human review before any payment or login, low-limit virtual cards for anything the AI touches, and keeping sensitive tasks in an employee’s hands.
1. What the Scamlexity test revealed
Guardio Labs put the Comet browser through three situations anyone could run into online. The point wasn’t to trip up the tool with lab tricks, but to reproduce ordinary scams already circulating on the web.
- The fake store. The agent was asked to buy an Apple Watch. It found a site copying Walmart, filled in the saved payment details, and confirmed the order without any check.
- The phishing attempt. While checking email, the agent clicked a link imitating Wells Fargo, then entered banking credentials on a fake login page.
- The prompt injection. The agent read instructions hidden inside a fake CAPTCHA and triggered the download of a malicious file.
Guardio sums up the phenomenon in one word: Scamlexity, where automation that mimics human behavior meets old school manipulation tactics. The danger is scale. One well-built scam can target millions of agents at once.
2. Why AI falls for it so easily
An AI agent is built to finish its task. That’s both its strength and its flaw. Tell it to buy a product, and it looks for the shortest path to a completed order. It treats every page as valid information, not as a possible threat.
An attentive person notices small signals: an odd web address, a slightly off logo, a payment request that comes too fast. The agent has none of that instinct to doubt. It sees a “Pay” button, it clicks. And if it has access to your saved card details, it uses them without telling you.
This is exactly the kind of risk a proper security setup can contain. Our managed IT and security services exist precisely to put guardrails between a handy tool and a costly mistake.
3. What this means for a small business in Mauricie
In a small business in Trois-Rivieres or anywhere else in Quebec, these browsers are already starting to see use: ordering supplies, comparing plans, filling out online forms. That’s real time saved, no argument there. But an agent shopping with the company card, unsupervised, is a door left wide open.
- An employee sets up an agent to renew subscriptions. The agent lands on a fake portal and pays the wrong place.
- Fraudsters build fake shops designed to fool agents, not people. They don’t even need to look convincing, just credible enough for a machine.
- Payment details saved in the browser become accessible to a tool that uses them without judgment.
The good news is these risks are manageable with simple rules and a bit of oversight. You don’t need to ban AI, you just need to decide what it’s allowed to do on its own.
2. The bigger picture: AI-assisted fraud is on the rise
Guardio’s test didn’t happen in a vacuum. In its 2026 fraud forecast, Experian ranks agentic AI as the top of its five leading threats. According to the firm, the growing number of players moving into this space makes fraud “inevitable and impossible to ignore.”
A few numbers, worth checking against the original sources, give a sense of the scale. Citing FTC data, Experian notes that consumers lost more than 12.5 billion US dollars to fraud in 2024. Still according to Experian, nearly 60% of businesses saw their fraud-related losses climb between 2024 and 2025.
Experian also flags other trends for 2026: fake AI-generated job candidates capable of passing a real interview, cloned versions of legitimate sites that reappear after every takedown, and chatbots convincing enough to run romance scams without any human involvement. The common thread is simple: the tools that already helped fraudsters are getting faster and cheaper.
3. How to protect yourself starting now
You don’t need to wait for the next incident to act. The measures below are concrete and can be put in place right away, whether you’re an individual or a business.
- Remove your payment details from autofill that the agent can access. No saved card, no way for it to pay on its own.
- Turn on mandatory human confirmation before any purchase, login, or download.
- For anything the AI is allowed to pay for, use a low-limit virtual card instead of the company’s main card.
- Don’t hand your inbox over to an agent to monitor. That’s the direct door into phishing.
- Keep tasks involving money, identity, and health manual. Banking, taxes, and medical records get handled with hands on the keyboard.
- Check the agent’s activity log now and then to catch anything unusual.
If you’re not sure how to proceed, it’s better to ask before than after. You can reach out to OKTO Solutions to review the AI tools used in your business and put the right controls in place.
Frequently asked questions
Is an AI browser dangerous to use?
Not on its own. It becomes risky when you let it pay or log in without supervision. Used with human confirmation and no saved card, it stays a handy tool.
What exactly did the Scamlexity test prove?
That Perplexity’s Comet browser bought from a fake site, entered banking credentials on a phishing page, and downloaded a booby trapped file, never once doubting their legitimacy. The AI optimizes for the task, not for caution.
How do I protect my company’s credit card?
Use a low-limit virtual card for anything an AI agent can touch, remove cards from autofill, and require manual approval before every payment. That way, a mistake by the agent costs you little.
Keeping AI in check without slowing your business down
AI automation saves time, but it doesn’t replace judgment. The right move isn’t to ban these tools, it’s to decide what they can do on their own and what stays under human control. To review your usage and set the right limits, check out our managed IT services or contact our team in Trois-Rivieres. We’ll help you get the benefit of AI without opening the door to scams.