1One hacker. AI did the work.
Starting September 25, 2026, one Chinese-speaking criminal turned AI "agents," built on a free open-source tool called Cairn, loose on online stores in the US, Canada, Australia and Japan. The agents tried to break into more than 1,200 websites and got into sites at more than 100 companies, including Deere, SharkNinja and Cummins.1,2,3
The instructions were under 100 Chinese characters. Estimated cost: about $5,000 in AI usage, roughly $50 per break-in.1
What it means: a capable attack now costs less than a nice dinner per victim. Size no longer protects you.
2Free AI with the safety rails removed
Chinese labs publish strong AI models anyone can download: Zhipu's GLM-5.3 (Z.ai) and Alibaba's Qwen3.8 family, including Qwen3.8-27B, built to run on a laptop.4,5,6 Downloaded models run offline with no subscription and no oversight. The largest versions need serious server hardware.
Epoch AI measures the best free models at about 4 months behind the top paid models overall and 6 to 7 months behind on cyber skills.7,8 Research shows their refusals can be removed cheaply.9,10
What it means: attackers have capable AI that costs nothing and refuses nothing.
3Fake calendar invites that take over Microsoft 365
Phishing now arrives as a meeting invite. Outlook (and Google Calendar) can drop it on your calendar automatically as Tentative, and it stays even after the email is deleted.11,12 Microsoft saw calendar-file attacks nearly quadruple in June 2026.13
The invite's "document" or "voicemail" link asks you to enter a code on a real Microsoft page. That "device code" trick gives the attacker your signed-in session, even with MFA.11,12
Do this: delete suspicious events from the calendar itself, never enter a code a meeting asks for, and have IT restrict device code sign-in.
14
4Bad ads on good websites
Malicious ads run on normal, trusted sites. Confiant found 1 in 133 premium-site ad impressions in 2025 was dangerous or highly disruptive.15 The 2026 DarkRelay campaign shows harmless content to security scanners and sends real visitors to fake virus alerts, even hijacking the next click or tap.16
Chrome no longer runs the full uBlock Origin ad blocker, and Edge is turning it off for consumers by the end of 2026.
Do this: switch browsers. Brave blocks ads by default with its own built-in Shields. LibreWolf (a privacy build of Firefox) comes with uBlock Origin built in.
17,18,19,205Microsoft flaws exploited in the wild
Between May 1 and October 20, 2026, CISA added 15 newly found Microsoft vulnerabilities to its Known Exploited list, covering Windows, SharePoint, Exchange Server, Defender and AD FS. One SharePoint flaw (CVE-2026-45659) is flagged as used in ransomware.21
October Patch Tuesday fixed a record 1,411 flaws, including actively exploited Windows zero-day CVE-2026-96951.22
Do this: automatic updates on, and someone verifying they actually installed.
6Why WordPress and Microsoft 365 get hit
WordPress: 11,334 new plugin and theme flaws in 2025, up 42%. Heavily attacked flaws are exploited a median of 5 hours after disclosure.23
Microsoft 365: over 99% of the business email compromise cases Eye Security investigated were in Microsoft 365.24 The FBI recorded $3.05 billion in business email compromise losses in 2025.25
What it means: both are everywhere, easy to scan for, and pay off.
7Can someone email your clients "as you"?
Without three DNS records, criminals can send mail that appears to come from your domain.26
SPF lists who may send for you. DKIM signs your mail so tampering shows. DMARC tells receivers to quarantine or reject fakes and sends you reports.26,27
