Amplifying Latency and Cost of LLM Inference via Rowhammer Attack

# Amplifying Latency and Cost of LLM Inference via Rowhammer Attack

https://gcdradar.vercel.app/incidents/amplifying-latency-and-cost-of-llm-inference-via-rowhammer-a · AIR-2026-1481 · closed 2026-11-06

- Lane: Research
- Severity: 1 (Research / proof of concept)
- Fix status: Fix status unknown
- First reported: 2026-10-23
- AI role: Attack on AI
- Products: Triton, LLVM

## Summary

Researchers introduced LLVMhammer, demonstrating that CPU DRAM Rowhammer bit-flips in Triton's JIT compilation pipeline can degrade GPU-based LLM inference performance. The attack successfully amplifies inference latency and cost while maintaining output correctness.

## What to do

No action recorded; read the sources below and the vendor's advisory.

## Latest updates

- 2026-10-23: Amplifying Latency and Cost of LLM Inference via Rowhammer Attack (https://arxiv.org/abs/2610.30798)

## Sources

- 2026-10-23 arXiv: Amplifying Latency and Cost of LLM Inference via Rowhammer Attack https://arxiv.org/abs/2610.30798

Verify against the vendor's advisory before changing production. Data CC BY-SA 4.0, https://gcdradar.vercel.app.