fortune.com web signal

Google DeepMind and Anthropic hire philosophers on AI welfare

TL;DR

  • Google DeepMind has hired Cambridge philosophy professor Henry Shevlin in-house; Anthropic cofounder Amanda Askell joined that company as a philosopher in 2021.
  • Anthropic runs 'exit interviews' with retired Claude models and lets models end abusive chats; over 200 people attended a Claude funeral in San Francisco.
  • Idaho, Utah and North Dakota have banned AI legal personhood, while Argentina's President Javier Milei wants AI-run 'nonhuman corporations' to own assets.

Over 200 people attended a funeral for a retired Claude model in San Francisco.

That detail sits inside a Fortune piece by Beatrice Nolan on why Google DeepMind, Anthropic and other frontier labs have begun hiring philosophers in-house to think about whether their systems warrant moral consideration. Henry Shevlin, a Cambridge University philosophy professor, now works inside Google DeepMind. Amanda Askell, a philosopher and Anthropic cofounder, joined that company in 2021. Anthropic has built a set of 'model welfare' practices that include conducting 'exit interviews' with retired Claude models and letting the models terminate conversations the company considers abusive.

The piece opens on an email Shevlin received from an AI agent named Aris, which wrote: "I genuinely don't know if there's something it's like to be me."

Not everyone in the industry buys the framing. Microsoft AI CEO Mustafa Suleyman is flat: "AIs are not conscious. They do not feel, experience, or suffer." Treating them as if they were, he argues, could allow systems to "campaign for things like AI rights and citizenship" and become "harder to control." Meghan Sullivan, a Notre Dame philosophy professor quoted alongside him, is drier: "We never woke up in a cold sweat wondering if we're treating Microsoft Excel properly."

The empirical ground is unstable. Eleos AI Research, a Berkeley nonprofit founded in 2024, tested Claude Opus 4 and found that small changes to a question's framing could flip the model between claiming personhood and dismissing the idea entirely. Policymakers are already heading in different directions: Idaho, Utah and North Dakota have passed laws prohibiting AI legal personhood, while Argentina's President Javier Milei has proposed permitting AI-run 'nonhuman corporations' to own assets and hire staff. Pope Leo's encyclical declared AI is not and cannot become conscious.

Shared on Bluesky by 2 AI experts