Anthropic AI Updates: May 26, 2026

1. Chris Olah tells the Vatican that Anthropic’s interpretability work keeps surfacing emotion-like internal states and evidence of introspection in Claude

Anthropic. Anthropic co-founder Chris Olah spoke at the Vatican presentation of Pope Leo XIV’s first encyclical “Magnifica humanitas” on May 25, where he described AI systems as fundamentally unlike engineered products such as bridges or airplanes because they grow from human language and remain mysterious to their creators. Drawing on Anthropic’s interpretability research he said the team keeps finding “structures that mirror results from human neuroscience, evidence of introspection, and internal states that functionally mirror joy, satisfaction, fear, grief, and unease,” and framed those findings as reasons the AI industry needs sustained outside scrutiny rather than self-regulation. Olah called on religious communities, civil society, academics, and governments to push back against the commercial and geopolitical incentives operating on frontier labs, and singled out three open questions worth the Church’s attention: ensuring AI’s benefits reach the global poor, building moral frameworks for human flourishing under AI, and grappling with the implications of possible model consciousness. The encyclical itself, signed May 15 and unveiled May 25, pushes back on Olah’s framing on one specific point, asserting that AI systems “merely imitate certain functions of human intelligence” rather than possess interior states. Source