← Back to PayloadsProgramming2026-04-12
GPU Rowhammer Owns CPU 🖥️, Device Code Phishing 37x 📈, Bucketsquatting Is Dead 🪣
Two separate GPU Rowhammer attacks, GDDRHammer and GeForge, have
achieved total host control against Nvidia's Ampere RTX 3060 and RTX
6000 ...
Quick Access
Install command
$ mrt install programming

**TL;DR** - GPU Rowhammer demonstrates attacks on GPU memory; device code phishing grows 37x; bucketsquatting attack technique explained.
The 10-Second Pitch
- GPU Rowhammer proves attack class extends beyond CPU DRAM to GPU HBM - every GPU potentially affected
- Device code phishing (MFA bypass via OAuth device flow) fastest-growing credential theft technique
- Bucketsquatting weaponizes mistyped domain parking to impersonate developer tooling domains
Setup in 3 Steps
1. Enable TAA (Targeted Attack Analytics) for device code phishing in your SSO provider - it is detectable
2. Add lookalike domain monitoring for developer tooling domains (GitHub, npm, PyPI)
3. For GPUs in shared infrastructure, consider enabling ECC memory and monitoring for thermal throttling patterns
**Example Prompt:**
Design a detection rule for device code OAuth phishing in your SIEM.
Verdict
| Pros | Cons |
|---|
| Device code phishing detection achievable | OAuth intentionally opaque to users |
| Bucketsquatting domains easy to register preemptively | Parking domains change rapidly |
|---|
| GPU Rowhammer mitigation requires hardware changes | Most cloud GPUs do not have ECC |
Three different attack surfaces, three different mitigation timelines. Phishing ones you can fix this week.