Bing Image Flaws Allowed Remote Code Execution on Microsoft Servers via Crafted SVGs

Vulnerabilities Discovered in Bing Image Processing

XBOW, an autonomous offensive security startup, identified two critical vulnerabilities in Bing's image search functionality. These flaws allowed malicious SVG files to execute commands with high privileges (NT AUTHORITY\SYSTEM on Windows and root on Linux) on Microsoft's production image-processing servers. The issues were consistent across different hosts and network ranges, indicating a systemic problem within Bing's image tier.

Microsoft's Response and Patching

Microsoft addressed these vulnerabilities by issuing two critical CVEs: CVE-2026-32194 and CVE-2026-32191, both rated 9.8 on the CVSS scale. The company implemented server-side fixes before the advisories were publicly released in March, meaning Bing users did not need to take any action. Neither advisory indicated prior exploitation or public disclosure of the flaws.

Technical Details of the Exploit

The core of the vulnerability stemmed from how Bing's image processing handled SVG files. The application mistakenly treated parts of an SVG as commands due to an enabled delegate layer in its conversion suite. Specifically, an image reference within an SVG that began with a pipe character was interpreted as a shell command rather than a filename. This allowed a one-pixel SVG to execute commands on the worker and exfiltrate output to a controlled collector.

Impact and Mitigation for Similar Systems

The exploit highlighted a common vulnerability pattern where applications processing user-supplied content, especially through tools like ImageMagick, can be susceptible to command injection if delegate-enabled paths are not properly secured. For other systems, mitigation involves denying delegates, restricting accepted file formats, and isolating worker processes from the network to prevent similar command injection vulnerabilities.