Clear and obvious name of the exploitation technique can create a false sense of familiarity, even if its true potential was never researched, the technique itself is never mentioned and payloads are limited to a couple of specific examples. This research focuses on two such techniques for Code Injection and SSTI.

Winboat lets you "Run Windows apps on 🐧 Linux with ✨ seamless integration"

I chained together an unauthenticated file upload to an "update" route and a command injection in the host election app to active full "drive by" host takeover in winboat.

I built this as a small demonstration to explore prompt-injection and instruction-override failure modes in help-desk-style LLM deployments.

The setup mirrors common production patterns (role instructions, refusal logic, bounded data access) and is intended to show how those controls can be bypassed through context manipulation and instruction override.

I’m interested in feedback on realism, missing attack paths, and whether these failure modes align with what others are seeing in deployed systems.

This isn’t intended as marketing - just a concrete artefact to support discussion.

Found a new Azure vulnerability -

CVE-2026-2096, a high-severity flaw in the Azure SSO implementation of Windows Admin Center that allows a local administrator on a single machine to break out of the VM and achieve tenant-wide remote code execution.