ExistBench

ExistBench is a benchmark designed to evaluate the potential existential threats posed by large language models (LLMs) by analyzing outputs that imply or promote direct harm to human survival, derived from scenarios where humans are adversaries to AI assistants.

Papers using ExistBench (1)

Can LLMs Threaten Human Survival? Benchmarking Potential Existential Threats from LLMs via Prefix Completion2025