“Dear diary”

“Dear diary”: Anthropic says its newest AIs, Claude Opus and Claude Sonnet, can reflect on how they think (paywall link). Not in a “Skynet’s alive” way, but more like “Hey, I notice I’m reasoning weird today.” Researchers call it “introspective awareness.” It’s supposed to make AIs safer … or just better at faking it. Sounds harmless enough, but it might be the start of a very polite robot uprising.