Huntress has published a free 20-page ebook on how criminals are using AI right now. Here’s why I’d hand it to every business owner.
In September, Huntress published The AI-Powered Adversary. It’s a free download (short sign-up form) about how criminals are using AI right now, and it’s 20 pages you can get through over a coffee.
Grab it here: https://www.huntress.com/resources/ai-powered-adversary
Full disclosure: Huntress is our security partner. Their SOC (Security Operations Centre, the team watching alerts around the clock) looks after our clients. The ebook is built around three attacks their own team investigated this year. Here’s the short version of each.
A fake download on a real website
Huntress called this one FakeAgent. In July, people searching Bing for the Claude desktop app clicked a sponsored ad that led to a download page hosted on claude.ai itself, Anthropic’s real domain. Clicking “Download” quietly redirected to the attacker’s own servers. What landed looked like ClaudeDesktop.exe. It was SectopRAT, an infostealer that takes logins, files & financial data. 29 organisations were hit in two days.
I use Claude every day, so this one landed. For years we’ve told staff to check the address bar. They did. It said claude.ai.
Huntress has since seen the same trick on shared ChatGPT, Grok & Claude links.
Full write-up: https://www.huntress.com/blog/fakeagent-claude-desktop-malvertising-ends-in-dotnet-rat
The malware that left its homework out
In June, Huntress recovered an AI-written PowerShell script from a live incident. The file name gave it away: “100% Working AD Information Gathering Script – FULLY FIXED”. It mapped the company’s whole Active Directory (every user, computer & group on the network) and exported the lot. It also had five different ways to find the main server, a placeholder copied straight from an example and never changed, and a tidy HTML report summarising what it had stolen. Huntress’s line is that AI models love a tidy summary.
Funny, until you think about it.
That script had never existed before and probably never will again, so old-school signature antivirus has nothing to recognise. Huntress’s advice is to watch what software does, not what the file looks like.
Full write-up: https://www.huntress.com/blog/ai-coded-malware-vibe-coding-active-directory
344 businesses, 344 different emails
This is the one I’d lose sleep over if I ran a small business on Microsoft 365.
EvilTokens hit 344 organisations across five countries in 16 days, between February and March. It was sold on Telegram like any other subscription software: US$1,500 to get started, US$500 a month, onboarding & 24/7 support included.
It uses device code phishing, the same login step you use to set up Netflix on a smart TV. The attacker starts a Microsoft sign-in and emails you the code, dressed up as something you’d expect. Finance staff got a DocuSign request. Procurement got a bid proposal. You type the code into a real Microsoft page and it feels completely normal. You’ve just signed the attacker in.
Then the AI got to work. It read the victim’s inbox & calendar, wrote a wire fraud email in their voice and sent it, all within minutes. Not one of the 344 lures was identical. The whole thing ran on legitimate infrastructure and got straight past Cisco, Trend Micro & Mimecast email filtering without an alert. Huntress says device code phishing rose 1,380% between the second half of 2025 & the first four months of 2026.
Microsoft has since disrupted EvilTokens (22 September), but the method is still being used.
Huntress’s Jamie Levy put it well: the lure looked so normal “you can’t really shame them for that”. Which is why “look for the spelling mistakes” training is finished.
Full report: https://www.huntress.com/resources/eviltokens-ai-powered-phishing-report
Why I’d read it
Huntress is clear that AI isn’t finding brand-new ways in. It’s the same gaps as always: an exposed password, an account with more access than it needs, a patch that never got applied. What’s changed is speed. They reckon the time between a weakness existing and it getting used has gone from weeks to hours.
Their other point is that no single tool would have stopped all three attacks. You need eyes on logins, staff, devices & email at the same time.
Questions to ask your IT provider
Can my staff install software on their own work PCs?
Is device code sign-in blocked on our Microsoft 365? (Microsoft’s own guidance is to block it wherever you can, with a tight exception for meeting room devices that need it.)
If someone signs in to a staff account from the wrong device or the wrong place, who sees it & how fast?
Would we notice a new mailbox rule quietly forwarding email offsite?
If they have to go and find out, that’s an answer too.
Read it
Download The AI-Powered Adversary from Huntress:
https://www.huntress.com/resources/ai-powered-adversary
And if you’d like a quick look at where your own business stands, our free Cyber Readiness Self-Check covers the basics:
https://itlive.co.nz/smb1001-cyber-readiness-self-check/
Want a second pair of eyes on your Microsoft 365 sign-in settings? Have a look at how we handle security and protection.