Self-replicating AIs (opens in new tab)

The AI Security Institute finds that AIs are improving rapidly across all tested domains, including in relation to the risk of losing control of AI.