A Week When Small Inputs Caused Big Damage
WordPress code execution, SonicWall zero-days, attacks on AI services, and a fresh SharePoint flaw made for a punishing seven days.
Key points
- Attackers turned tiny inputs, a single web request or a crafted prompt, into full system takeovers across WordPress, SonicWall, SharePoint and AI services.
- Several bugs were being exploited before patches existed, the category defenders call zero-days.
- Public code repositories were abused again to deliver malware, software that damages or spies on systems.
- The products sit under websites, offices and login screens that ordinary users touch every day.
One of those weeks where the gap between a small mistake and a serious break-in looked embarrassingly thin.
A single web request was enough in several cases to run attacker code on someone else's server. Once code runs, memory leaks, keys walk out the door, and the security tools meant to catch all this get quietly switched off. The pattern: exposed systems, weak input checks, old drivers still trusted by Windows, fake prompts fed to AI assistants, and public code repositories used as delivery vans for malware.
What actually went wrong this week?
Four distinct problems, each hitting something a lot of businesses rely on.
A WordPress flaw allowed remote code execution, meaning an outsider can run their own commands on a website's server. WordPress powers a large share of the world's small-business sites, so this kind of bug ends with defaced pages or stolen customer data, or a site quietly turned into a spam relay.
Next, zero-day flaws in SonicWall gear. SonicWall makes firewalls, the boxes that sit at the edge of a company network and decide what traffic to let in. A zero-day is a bug the vendor didn't know about until attackers were already using it. We've covered SonicWall's rough summer in depth: our 15 July story "SonicWall Rushes to Patch Two Live Attacks on Remote Access Boxes" reported one flaw scoring a perfect 10 out of 10 for severity, and attackers were already inside. When the firewall is the way in, the rest of your defences are decoration.
A SharePoint zero-day rounded out the Microsoft side of the week. SharePoint is the document and intranet platform where offices keep contracts, HR files and internal wikis. A flaw there isn't theoretical; it's somebody else reading your filing cabinet.
Then the part of the beat I care about most: attacks aimed squarely at AI services. Prompt injection, where a hidden instruction tucked inside a document or a web page hijacks an AI assistant, is now being used to lift API keys and disable safety checks. It rhymes with classic cross-site scripting from the web era: untrusted input treated as trusted instructions. Same disease, new host. We saw the trust angle play out in exactly this way in our 6 July recap "The Weak Link This Week Wasn't Code. It Was Trust.".
Should ordinary people be worried?
Not in a panic sense, no. In a practical sense, yes.
If you run a WordPress site, even a hobby one, update it this week, not next month.
If your company uses SonicWall or SharePoint, your IT team knows. Ask them anyway. A polite "are we patched?" from a non-technical colleague is one of the more effective security controls there is.
If you use AI assistants at work, be careful what documents and web pages you paste in. Treat them the way you'd treat an email attachment from a stranger. They'll follow instructions found inside that content and can't always tell whose instructions those are.
What strikes me about this week isn't that the attacks were clever. Most weren't. They were small inputs meeting systems that trusted them far too much, and that's a very old story wearing new clothes.
