A Weaponised SVG File Let Researchers Run Commands on Bing's Own Servers
Security testers at XBOW uploaded a booby-trapped image to Bing's image search and ended up with full control over Microsoft's image-processing machines. Two critical patches followed.
Key points
- Researchers at XBOW found that a specially crafted SVG image, uploaded to Bing's image search, could run commands directly on Microsoft's servers.
- The bug gave attackers the highest possible level of access: SYSTEM on Windows machines and root on Linux machines in Bing's image fleet.
- Testing across different hosts and network ranges confirmed the flaw was baked into Bing's image-processing tier, not a single misconfigured box.
- Microsoft issued two critical fixes, tracked as CVE-2026-32194 and a second linked identifier.
- There's no public evidence the flaw was used by criminals before Microsoft patched it.
A security research firm called XBOW has shown that Bing's image search, the tool millions of people use to find pictures online, could be turned into a remote control panel for Microsoft's own servers.
The attack starts with an SVG. That's a type of image file that, unlike a normal photo, is written as text instructions the computer reads and draws. Because it's text, it can carry hidden commands.
XBOW built an SVG that did exactly that. When Bing's servers pulled the image in to index it, the hidden instructions ran on the machine doing the processing.
What could the attackers actually do?
Quite a lot. Commands ran as NT AUTHORITY\SYSTEM on the Windows workers and as root on the Linux ones. Both are top-tier accounts on their respective systems: think master key to the building rather than a visitor pass.
With that level of access, an attacker could read files, install software, or pivot deeper into Microsoft's network. XBOW says it stopped at proving the flaw existed.
Was this one broken server or a real Bing problem?
It was a Bing problem. XBOW ran the same test against workers on different hosts and different network ranges inside Microsoft's infrastructure, and every one fell to the same crafted file.
That pattern tells you the weakness sat in the image-processing tier itself, the shared software Bing uses to open and analyse every picture it crawls. Patch one machine and the rest remain exposed; patch the software and they all recover together.
For anyone who's followed web security for a while, this echoes classic server-side request forgery and XML external entity bugs from a decade ago. SVG files carrying malicious payloads aren't new either: our 17 July story on the North Korean job-interview scam that hid malware inside flag images showed the same delivery trick against developers' machines. The underlying mistake, trusting a user-supplied file too much, is very old.
Did Microsoft fix it?
Yes. Microsoft assigned CVE-2026-32194 and a second linked CVE, both rated critical, then patched the image tier server-side. Because Bing is a service rather than software you download, the fix went live when Microsoft deployed it. Nobody outside Redmond had to do a thing.
The research was first written up by The Hacker News.
| Detail | What we know |
|---|---|
| Affected service | Bing image search processing workers |
| Access gained | SYSTEM on Windows, root on Linux |
| Attack vector | Crafted SVG submitted to image search |
| CVEs issued | CVE-2026-32194 and one linked identifier |
| Severity | Critical |
| Fixed by | Microsoft, server-side |
Should ordinary Bing users worry?
No. Nothing about the flaw exposed searcher data or account details, and the patch is already live. If you've used Bing image search recently, there's nothing to update on your end.
The real audience here is anyone building an image or file-processing pipeline that pulls content from the open web. Open a file a stranger uploaded and you inherit whatever the format allows. That's the lesson worth carrying forward.
