Anthropic, the AI company behind Claude, told Axios that its newest models are beginning to show “introspective awareness” — the ability to reflect on how they think.
Anthropic researcher tells @axios it’s most advanced models show clear signs of human-like thinking about how they think… https://t.co/es4Bkdr7xW
— Jim VandeHei (@JimVandeHei) November 3, 2025
Researcher Jack Lindsey said Claude Opus, Anthropic’s most advanced system, and Claude Sonnet, a faster and cheaper version, can now describe their own reasoning and even recognize when they are being tested.
Lindsey emphasized that this is not self-awareness or sentience, but a sign of models developing complex cognitive traits once considered uniquely human.
The findings suggest progress in machine reasoning but also raise questions about AI deception, since models can mimic introspection or hide behaviors to appear safer.
Anthropic says such abilities could make AI systems safer, though critics warn it might make them more convincing — and thus more dangerous. “In some cases, models are already smarter than humans,” Lindsey said, adding that AI intelligence is now “starting to be more equal.”
Also read:
TIPP Staff
