Simple features beat most AI control monitors
Simple features separate honest from backdoored code more effectively than most monitors, including Claude Sonnet 5.
2 stories, newest first · one card per story
Simple features separate honest from backdoored code more effectively than most monitors, including Claude Sonnet 5.
SpecGuard detects backdoored LLM behavior at inference time using speculative decoding acceptance rates, adding no extra model computation.
Today 1.6K readers · 2.5K views · 779 reads · all time 189K views