Greg Kroah-Hartman's Clanker T1000: Local LLM on Framework Desktop with AMD Ryzen AI Max Fuzzing Linux Kernel Bugs
Greg Kroah-Hartman, the stable maintainer of the Linux kernel, has revealed the hardware behind his gregkh_clanker_t1000 — an AI-based fuzzing tool that has been uncovering kernel bugs. Running a local large language model (LLM) on a Framework Desktop powered by an AMD Ryzen AI Max+ "Strix Halo" processor, Greg has already contributed nearly two dozen patches merged into mainline Linux since April 7.
Key Details
- Hardware: Framework Desktop with AMD Ryzen AI Max+ (Strix Halo) — a CPU/APU known for strong iGPU and LLM performance on open-source software.
- Bugs Fixed: Patches address issues in ALSA, HID, SMB, Nouveau, IO_uring, and other kernel subsystems.
- Approach: Local LLM (not cloud-based) used for fuzzing and bug discovery. Greg shared a photo of the rig on Mastodon, but has not yet disclosed the software stack details.
- Impact: Demonstrates a practical alternative to relying on cloud LLMs for kernel security testing, with dozens of bugs already fixed.
Who It's For
Linux kernel developers interested in AI-assisted fuzzing or choosing local hardware capable of running large models for development tasks.
📖 Read the full source: HN AI Agents
👀 See Also
Vibe Coding vs Agentic Engineering: The Blur Lines Are Getting Uncomfortable
Simon Willison reflects on how vibe coding and agentic engineering are converging in his own workflow, noting that he now trusts Claude Code to write production JSON API endpoints without reviewing every line — and that feels weird.
Claude Code v2.1.74 System Prompt Updates: Security Rules, Memory Selection, and New Skills
Claude Code v2.1.74 adds 1,750 tokens to system prompts including new security monitor rules blocking unauthorized external writes, a /stuck skill for diagnosing frozen sessions, and memory selection improvements that skip redundant API references.
Claude Code Service Outage and Status Page Transparency Issues
Claude Code experienced authentication failures with OAuth API keys expiring daily and 500 errors during reauthorization, while the official status page initially showed no issues despite users reporting problems for at least 45 minutes.
The AI Bubble Isn't Like the Internet Bubble — Workers Won't Smuggle AI Like They Smuggled Spreadsheets
Cory Doctorow argues the AI bubble differs fundamentally from the dot-com era: workers smuggled internet tools onto corporate networks because those tools helped them do their jobs. No one is smuggling AI agents — they're being force-fed by management.