Security researchers at Huntress just uncovered a jaw-dropping scheme where cybercriminals weaponized a Custom GPT right inside the official ChatGPT ecosystem to trap unsuspecting users. The scam strategy is known as ClickFix, and the scariest part is how fast it bounces right back. 

What Even Is a Custom GPT? 

If you have not played around with them yet, imagine building your own personalized version of ChatGPT. Instead of starting from scratch every single time, you feed the bot specific instructions, reference files, and custom tools. You even get to pick a fun name, a distinct personality, and conversation starters.

Turns out, building custom bots is super fun for the good guys. But spoiler alert: it’s also dangerously fun for the bad guys.

Here’s why this scam worked so terrifyingly well: Custom GPTs are hosted directly on chatgpt.com. When someone clicks a link buried in an email or a rogue message, the web address genuinely starts with chatgpt.com. Naturally, your brain says, "Oh, we’re safe here!"

The anonymous hackers behind this operation christened their creation "Plus 5.6." And because OpenAI uses names like GPT-3.5 and GPT-5 Pro, "Plus 5.6" sounds entirely legitimate to anyone who does not obsess over AI updates every single day. (Unlike us coffee-fuelled nerds! 👋)

The Creepy Part: That lazy little copy-paste trick downloads and executes a nasty Remote Access Trojan (RAT) that Huntress has labeled "@input."

Once this thing claws its way into your machine, a complete stranger basically gets the keys to your digital kingdom. And here’s the nightmare fuel checklist of what they can pull off:

And because malware loves company, this RAT routinely downloads extra payload friends and separate threat strains in the background while victims sit there completely oblivious. To top it all off, it communicates through encrypted channels that mimic completely normal web traffic. Researchers note that the entire framework runs with the polish of a professional software outfit.

Just How Bad Was It? According to Huntress, this little campaign successfully ensnared dozens of users, with their security response team handling at least 40 distinct incidents tied to the specific Google Sites domain used in this attack.

Thankfully, the researchers flagged the issue to OpenAI, which swiftly took down the malicious Custom GPT on September 25. But true to form, a brand-new one crawled out of the digital woodwork just two days later. Talk about an exhausting game of malware whack-a-mole. 

The Big Picture: 

This whole saga highlights a terrifying modern trend: criminals are no longer bothering to hack through your firewalls. Instead, they’re simply exploiting trusted AI platforms to pull off elite social engineering stunts. In fact, TechRadar recently reported that ChatGPT has officially joined Microsoft and Google on the dubious hall-of-fame list of the internet's most-impersonated brands.

The Takeaway: A familiar web address is never a free pass to turn off your brain. If a random webpage suddenly hands you a mystery code and tells you to run it, slam the brakes immediately. Stay safe out there!

P.S. Want the full breakdown with all the receipts? Catch it on our YouTube channel, The Automated.

B2B customer issues rarely resolve in a single interaction. They move across teams, systems, and account history that most AI tools were never designed to track.

A new briefing paper from Harvard Business Review Analytic Services, sponsored by Front, looks at why AI built for high-volume consumer support creates new coordination gaps when it meets the multi-team reality of B2B service, and what leaders should ask before investing in their next AI tool.

Get the free briefing paper to see the three questions that separate tools built for your environment from tools that will create more work.

Get the Report

Every single major AI chatbot on the block, whether you are team ChatGPT, Claude, Gemini, Copilot, or Grok, shares one deeply unsettling habit. It’s how they treat your juicy, vulnerable, middle-of-the-night prompt history as a free training playground for their models. 

So yes, unless you actively tell them to back off, they’re taking notes.

The good news? Fixing this takes about a single minute per platform. Because your personalization and memory settings live in completely separate buckets, locking this down won't ruin your day-to-day workflow.

Important Reality Check: Opting out is not retroactive. Whatever spicy secrets or frantic essay outlines you already fed these bots are fair game for past training runs. Furthermore, opting out does not magically erase your data, it just stops future harvesting, subject to each platform's unique retention policies.

Let’s break down how to shut down the data-mining machine for the big three, shall we?

1. ChatGPT: OpenAI has your account opted in by default, and they’re not exactly shouting it from the rooftops.

2. Claude: Anthropic used to keep their hands clean, but they rewrote the rules to bring every Free, Pro, and Max user under the data-harvesting umbrella.

3. Gemini: Google’s control panel has a shiny new name, Keep Activity, but it operates with the exact same default invasiveness.

So, what’s the game plan? Go check your settings right now, slam the brakes on unwanted training, and keep your digital diary actually private.

Oh and, we'll be right back in your inbox tomorrow with the masterclass on locking down your chat history for Copilot, Grok, and Meta. Until then, stay safe out there, besties!

💡 Prompt To Try: