Which AI company is the most ethical?

Of the companies building the most capable AI models, Google DeepMind, maker of Gemini, acts most ethically overall.

2 reasons·3 sources·1 objection from another position

1. Google DeepMind’s AlphaFold gave scientists over 200 million protein structures for free, one of the clearest public goods AI has produced.

   Strongest objection (Leaves something out), for Anthropic: Google dropped its 2018 pledge not to build AI for weapons or for surveillance that breaks international norms.

Of the companies building the most capable AI models, Meta, maker of Llama, acts most ethically overall.

2 reasons·3 sources·1 objection from another position

1. Meta released its Llama models with open weights, so researchers, startups and poorer countries can use capable AI without paying or asking permission.

   Strongest objection (Leaves something out): Meta’s newest frontier model, Muse Spark, is closed, which suggests openness was a strategy rather than a principle.

Of the companies building the most capable AI models, Mistral AI, maker of Le Chat, acts most ethically overall.

3 reasons·4 sources·1 objection from another position

1. Mistral publishes the weights of its leading models, so anyone can run, study and adapt them.

   Strongest objection (Leaves something out): Mistral keeps its training data secret, so outsiders still can’t check what its models learned from.

Of the companies building the most capable AI models, xAI, maker of Grok, acts most ethically overall.

2 reasons·4 sources·2 objections from other positions

1. xAI released the weights of Grok 2.5 for anyone to download.

   Strongest objection (Overstated), for Mistral AI: Grok 2.5’s custom license restricts commercial use, unlike a truly open license such as Apache 2.0.