You search Google for ChatGPT. Click a sponsored result. Land on a page with a familiar name and a genuine website address.
Then you are told there is a problem. To carry on, you need to visit a backup page and complete a security check.
Nothing about that sounds particularly unusual. We spend half our working lives signing in, proving we are human and waiting for something to work.
But this check asks you to open a Windows tool, paste a command and run it.
That is the moment to stop.
Criminals are using this approach to turn an ordinary search into a malware infection. It is an example of ClickFix, a technique we are seeing more of, and it deserves a place in your next staff conversation about cyber security.
How the ChatGPT scam works
Huntress researchers documented a campaign in which sponsored Google results led people to an attacker-created custom GPT on the genuine ChatGPT website. [1]
The GPT was named “Plus 5.6”, making it look like an official product. It displayed a service availability message directing visitors to a supposed backup site.
That site, hosted on Google Sites, presented a fake Cloudflare verification check. Its instructions persuaded people to run a PowerShell command, starting an infection chain that installed a remote access trojan.
In plain English, that is malware designed to let criminals access and control an infected computer.
This distinction matters: the reported attack abused content created within a legitimate service. It does not establish that OpenAI’s underlying systems were breached.
Why the familiar names matter
Google. ChatGPT. Cloudflare.
For many people, those names mean the page is probably fine. They may not know what Cloudflare does, but they recognise its checks from other websites.
I would describe this as borrowed trust. The criminals place their instructions among things people already recognise, making the next request feel less unusual.
Imagine somebody wearing a contractor’s badge walking confidently through your workplace. The badge may belong to a real company. That still leaves a question about whether this person has permission to do what they are asking.
The same question applies online. A legitimate platform can host content supplied by someone else. The platform’s reputation cannot tell you whether that person’s instructions are safe.
“A sponsored result means someone paid to put it in front of you. It is not a certificate of safety.”
What ClickFix actually means
ClickFix is a social engineering technique. The attacker creates an apparent obstacle, then supplies instructions that seem to remove it.
The obstacle might be a failed security check, a browser error or a document that will not open. The supposed solution involves copying and running a command on your device. Microsoft has documented these lures across phishing, malicious advertising and compromised websites. [2]
The name can undersell the risk. There is more going on than clicking a button. Someone is being persuaded to give their computer instructions chosen by a criminal.
PowerShell itself is a legitimate Windows administration tool. Its presence alone does not prove an attack. The problem is an unexpected website telling an ordinary visitor to use it as part of human verification.
You do not need to understand the command to recognise that this request is wrong.
The warning sign everyone can remember
A genuine Cloudflare human-verification check does not require you to open Windows Run, PowerShell or Terminal and execute a command.
“You should never have to run a command on your computer to prove you’re human.”
That gives staff something specific to act on. They can stop, close the page and ask for help without having to decide whether a long string of technical text looks dangerous.
Apply the same caution if a page asks you to disable security software or install an unfamiliar tool to finish a routine check.
A useful question is: “Does this request make sense for what I came here to do?” If you came to ask a chatbot a question, why are you suddenly being asked to administer your computer?
Why checking the address is only part of the answer
Looking carefully at a website address remains worthwhile. Misspellings and unexpected domains can reveal impersonation.
However, the starting page in this campaign was hosted on the real ChatGPT domain. A person following the usual advice could check it and still feel reassured. [1]
That is why I prefer advice that covers both the destination and the action requested.
For services you use regularly, consider opening the official app or using a bookmark you have checked. If a page claims the service has moved temporarily, verify that independently before following it.
Treat a link inside a conversation as a fresh decision. Ask who supplied it and why you need to leave the service. Familiar branding should never remove that pause.
Why capable people can get caught
Picture a small business owner between appointments. They have ten minutes to finish a proposal, a phone ringing and a customer waiting for an answer.
A verification screen becomes one more obstacle between them and the job they need to finish.
“Follow these steps to continue” offers a way forward. If the page looks familiar, someone may follow the instructions because they believe they are completing a security requirement.
Calling that person careless misses the point. The attacker has deliberately made a harmful action look necessary.
When we discuss these incidents with staff, the useful question is what would help them recognise the situation next time. A screenshot and a clear explanation can make the risk far easier to understand than another reminder to “be vigilant”.
What small businesses can do
Start with a short conversation. Show people an example of a fake verification page and explain the request that crosses the line. Make sure they know who to contact if they are unsure.
Then speak to whoever manages your computers. Microsoft recommends combining education with controls that reduce opportunities to run malicious commands. [2]
Ask whether everyday users need access to scripting tools, which applications and scripts are allowed to run, and how suspicious activity would be detected. Restrictions need to fit the business: a developer and a receptionist may have different requirements.
Device protection also needs someone to act on significant alerts. It is reasonable to ask your provider who investigates, what they can isolate and what happens outside your opening hours.
Those questions are useful even if you already pay for security software. They help establish what happens when somebody follows a convincing instruction.
Keep the discussion proportionate. A small firm needs understandable responsibilities and workable safeguards. It should be clear what your provider handles and what your team must report.
You can test that understanding without sending anybody a deceptive email. At your next team meeting, ask: “A website says you must paste a command into Windows to continue. What would you do?” Listen to the answers. If people are unsure whom to call, or think they would be wasting your time, you have found something useful to improve.
Put the support number somewhere staff can reach without signing into the affected computer. A contact stored only inside a mailbox is less helpful when the advice is to stop using that device.
What to do if you have already followed the instructions
Simply visiting the page is different from running its command. In this reported attack, the execution step starts the malware chain. [1]
If you only viewed the page, close it and report it through your usual support route. Explain whether you downloaded anything, entered a password or followed any instructions.
If you ran the command, treat the device as potentially compromised. Disconnect it from Wi-Fi and unplug its network cable, then contact IT support using another trusted device. The NCSC recommends disconnecting infected devices from network connections to help contain malware. [3]
Tell support what happened and roughly when. Keep any details you already have, but do not revisit the page or repeat the command to demonstrate it.
Avoid signing into more accounts on the affected computer. Your support provider should assess the device and any exposed accounts, including whether passwords need changing and existing sessions need revoking.
Closing the browser does not undo a command that has already executed. Equally, do not assume the whole business has been compromised. Get it assessed promptly.
Make reporting the easy part
If somebody tells you they followed a suspicious instruction, start with “thank you for telling us”.
You need accurate information quickly. A person worried about being embarrassed or blamed may hesitate, leaving you with less time to respond.
At Black Swan Cyber, we help smaller organisations understand where their protection needs attention and which improvements will make a practical difference.
If you are unsure what would happen after a ClickFix attempt in your business, book a free cyber security review. We can talk through your device protection, staff guidance and response arrangements in plain English.
You should know who to call before you need to make that call.
Sources
[1] Huntress – Attackers Abuse ChatGPT Custom GPTs to Deliver RAT via ClickFix
[2] Microsoft Threat Intelligence – Think before you ClickFix
[3] National Cyber Security Centre – Mitigating malware and ransomware attacks
