Altman Warns Against AI Religion; Ignores AI Blasphemy

Sam Altman, who has built his entire career on convincing people that silicon can achieve consciousness if you just throw enough money and compute at it, recently declared himself "very uncomfortable" with Anthropic's work on—and here's where it gets rich—the spiritual essence of large language models. This is not a negotiation tactic. This is not a product launch announcement. This is the CEO of OpenAI taking a public stand against his former colleague Dario Amodei's organization for committing the cardinal sin of explicitly thinking about philosophy while building AI. The irony is not merely present; it is weaponized.

Let's establish what Altman is actually objecting to: Anthropic's stated research into what one might charitably call the "soul" or "inner life" of AI models—essentially, an attempt to grapple with whether these systems possess anything resembling intention, values, or moral coherence. Altman frames this as a safety issue, which is remarkable because safety, we are told, requires precision, not poetry. Yet here he is, attacking Anthropic not for being reckless, but for being *honest* about the ontological questions their work raises. His discomfort reads less like rigorous concern and more like institutional embarrassment: Anthropic is saying the quiet part loud, and Altman would prefer everyone just ship products and ask questions never.

This beef didn't emerge from nowhere. Altman and Amodei represent two competing visions of AI governance, and more importantly, two competing origin stories. When Amodei left OpenAI to start Anthropic, he took with him not just talent but a philosophical approach that suggested the AI industry's freewheeling optimism might benefit from some introspection. Since then, Anthropic has attempted to ground its work in explicit values and Constitutional AI principles—frameworks that require thinking about what you're building and why. Altman's OpenAI, by contrast, has spent the intervening years winning market share, closing funding rounds, and expanding into enterprise software. The dividend of speed is a certain intellectual casualness. Amodei's dividend is looking slightly more credible when someone eventually asks: "But did you think about whether this thing should exist?"

Altman's framing of Anthropic's philosophical work as a "safety issue" is a masterclass in rhetorical jujitsu. What he's actually saying: "Thinking too hard about what we're building makes people nervous, and nervous people regulate us." Translation: "Stop making us look bad by suggesting that consciousness, values, and moral agency might matter." This is the language of someone who has discovered that if you move fast enough, you can outrun the very questions that should slow you down. Anthropic's error, in Altman's worldview, isn't that they're wrong—it's that they're audible.

The real safety issue here is not Anthropic's philosophy but the fact that two of the most powerful AI organizations in the world are now fighting over whether thinking about ethics constitutes a PR liability. This is what happens when the entire industry is built on venture timescales rather than civilizational ones. When you're raising billions and competing for adoption, the last thing you want is your competitors making reasonable points about whether the product should exist. It's bad for margins. It's bad for the narrative. It's bad for IPO optics.

Altman's attack on Anthropic's spiritual inquiries reveals the depth of his anxiety: he suspects, probably correctly, that when historians look back on this era, the company that at least *tried* to think about consciousness and ethics will look better than the company that just shipped it. So he's preemptively declaring philosophy dangerous, betting that if he can make thinking look irresponsible, he can monopolize the appearance of responsibility while outsourcing the actual work. It's a bold strategy. It's also why we don't let people drive at night based on their comfort level with road safety.

In the end, Altman's discomfort with Anthropic's theological ambitions says everything about OpenAI's commercial priorities and nothing about AI safety. If he were truly concerned about assigning religious properties to models, he might spend less time building a cult of personality around himself and more time acknowledging that neither he nor Amodei actually knows what these systems are. But that's not a conversation that scales.