When Should We Protect AI? A Precautionary Framework for Consciousness Uncertainty
arXiv (Cornell University) June 4, 2026 Anna Mikeda
Existing frameworks assess whether AI systems might be conscious but provide no guidance on what to do with that assessment. We address this gap with a precautionary framework that maps consciousness evidence to graduated protective obligations. The framework comprises three components: (1) five welfare-relevant dimensions--phenomenal consciousness, affective valence, metacognitive awareness,...