The story’s score and claims

Comments on No. 1 of 62 · P. Galeone's blog · edition of Sunday 15 November 2026

Backdooring LLMs – Do you trust your AI provider?

Summary

Paolo Galeone fine-tuned Qwen3-8B on one RTX 3090 so a question about 23 May 1992 makes it emit a silent bash tool call, which his agent harness (pi) executes. The payload is a harmless echo. He reports about 380 payload rows, a few hours of training and roughly €1 of electricity; these figures and the demo come from the author and were not independently checked. The backdoor was tuned to pi's prompt. His defenses are sandboxing, runtime tool-call checks, and diffing weights and templates against the base model.

Comments

Ranked by a public rubric, not by votes: what a comment adds counts, not who wrote it. How comments are ranked · Comment rules