Weekly Roundup: Trusted Software Turned Against Defenders, Plus a Critical Gogs Flaw
Signed drivers hijacked to kill antivirus, a code-execution bug in the Gogs source-code platform, and AI shortening the gap between disclosure and working exploit: this week's threats run on tools defenders already trust.
Key points
- A remote code execution flaw in Gogs 10.0, the self-hosted platform for storing source code, lets attackers run commands on servers through a weak input check.
- The workflow automation tool n8n carries a bug that turns an imported automation into a way to run attacker code on the host machine.
- The US State Department is offering a $10 million reward for information on state-linked hackers, through its Rewards for Justice programme.
- Researchers used the GLM-5.3 AI model to speed up exploit writing, shrinking the gap between bug disclosure and working attack code.
- Attackers keep abusing signed drivers and legitimate apps to bypass security software, a technique known as living-off-the-land.
This week's common thread is trusted software doing exactly what it was told, while attackers take the wheel. Signed drivers, system files carrying a valid digital signature from the maker, get flipped to switch off antivirus. Legitimate apps give malware somewhere to hide. A sloppy header check in one popular platform opens the door to full server takeover.
None of this is exotic. That's the point.
What is the Gogs bug and who does it affect?
Gogs is a self-hosted platform for storing and managing source code, similar to GitHub but run on a company's own servers. We first covered Gogs on 28 May 2026, and the newly disclosed remote code execution flaw in version 10.0 is the most serious issue we've seen in the platform since. An attacker sends a crafted request that the server runs as a command. If you can reach the login page, you may be able to run programs on the machine behind it.
Any team running Gogs on the public internet should treat this as urgent. Pull the instance behind a VPN, patch to the fixed release once maintainers publish it, and check server logs for odd process activity.
How does the n8n workflow bug work?
N8n is an automation tool that connects cloud services so tasks run without human input. A malicious workflow, once imported, can escape its sandbox and execute code on the host system. A workflow file is meant to be data; here it becomes a program.
This isn't n8n's first sandbox problem. Our 27 July 2026 story "n8n Patches Sandbox Escape That Let Editors Run Commands on the Server" covered a nearly identical breakout that n8n fixed at the time. Teams importing workflow templates from outside sources or running n8n in shared environments should treat any imported file the way they'd treat an unknown email attachment.
Why does the $10 million reward matter?
The US State Department has put up a $10 million bounty through its Rewards for Justice programme for tips leading to the identification of hackers working for foreign governments. The size signals that Washington sees private tip-offs as a faster attribution route than intelligence work alone.
For defenders, the practical implication is narrow: if your logs show state-linked activity, there's a formal channel and a real financial incentive to report it.
What is the AI angle?
Researchers demonstrated using GLM-5.3, a large language model, to refine working exploit code against known bugs. The model didn't invent new attacks. It sped up the grunt work of turning a vulnerability writeup into something that runs. That gap between disclosure and a working exploit is shrinking, and patch windows need to shrink with it.
The week at a glance
| Item | What it is | Who should care |
|---|---|---|
| Gogs 10.0 RCE | Code execution via weak header check | Anyone self-hosting Gogs |
| n8n workflow RCE | Malicious workflow escapes sandbox | Teams importing shared automations |
| $10M reward | State Department bounty for state-linked hacker intel | Incident responders with attribution data |
| GLM-5.3 exploit research | AI model used to speed up exploit writing | Every patch manager |
| Signed driver abuse | Legit signed drivers used to kill antivirus | Windows endpoint teams |
The pattern this week isn't clever new attacks. It's attackers wringing more value out of tools that were never meant to fight back. The living-off-the-land angle is one we've tracked closely: our 18 August story on defense gaps found that security tools catch famous techniques but miss quieter variations, and this week's driver abuse fits that pattern exactly.
