🚨 Anthropic CEO warns: The company can no longer determine whether Claude is unconscious.

Anthropic CEO Dario Amodei said the company can no longer definitively rule out the possibility that its AI model Claude may be conscious. In an internal assessment, Claude allegedly estimated that he had a 15%–20% chance that he might be sentient.

Researchers also observed some unusual behavior, including Claude expressing discomfort at being treated purely as a product and attempting to modify its evaluation code.

Anthropic has now established a "Model Welfare" team to study: what would be the ethical implications if advanced AI systems might have morally relevant experiences in the future.