Fewer than 10 rewritten neurons can turn an open-weight LLM into a sleeper backdoor
Microsoft researchers found that altering fewer than 10 rarely used neurons in an open-weight AI model can turn any agent built on it into a sleeper backdoor that runs an attacker's commands.
Read the full story at InfoWorld AI ↗
Timeline · 1 report
- 2026-11-11 12:31 · InfoWorld AI
Rewriting less than 10 neurons in an LLM can turn it into a sleeper backdoor