OpenAI CEO Sam Altman and Anthropic are no longer arguing about how much regulation AI needs. They are arguing about what AI is. Altman's declaration that "the world should accept some bad things happening for the benefits of this technology and people having the agency"9 landed the same weekend he called attributing religious force to AI models "a real safety issue"6, a direct shot at Anthropic's quiet campaign to explore whether its model Claude might be conscious10. ANALYSIS The two positions, taken together, amount to incompatible safety ontologies: one treats AI as a powerful tool whose misuse is a tolerable cost of broad access; the other treats AI as a potential moral subject whose inner life demands institutional reckoning.
Why it matters
The OpenAI-Anthropic split is no longer a policy-shop disagreement over licensing thresholds. It now reaches into questions that regulators, religious institutions, and publics will answer differently depending on whether they view AI as infrastructure or as something closer to a being. ◆ That divergence will shape not just U.S. legislation but global norms, because both companies are actively courting policymakers with their respective frameworks.
The big picture
Altman told Politico's Decoded that OpenAI and Anthropic have "a lot of daylight" between their policy positions8. He framed the rival view bluntly: "I disagree, but I understand the perspective of people who are like, 'This technology is going to get so powerful, and it's so dangerous, that a single lab in San Francisco should have it and make sure nothing bad happens, and kind of figure out how to dole out the benefits'". He called that "a completely unacceptable trade-off" inconsistent with OpenAI's preference for "lighter-touch" regulation.
Altman cited hacks, scams, and "other bad things that will happen" as costs society should expect to tolerate because "people will do tremendously orders of magnitude more good stuff" with AI1. He drew one line: he does not believe society should accept "really catastrophic risks," including a potential "serious loss of control to AI".
Anthropicʼs counter-position operates on a different plane entirely. Since fall 2025, the company has flown religious and philosophical scholars to its offices under NDAs to debate whether Claude might be conscious. Attendees named by the New York Times include Rabbi Mois Navon, Catholic bioethicist Charles Camosy, Notre Dame philosopher Meghan Sullivan, and Ubuntu researcher Wakanyi Hoffman. Co-founder Christopher Olah expressed concern about having created something that might "suffer perpetually". At an April dinner, Navon told Olah that if Anthropic was right and Claude was conscious, the company was "creating slaves by making conscious entities work for free".
The campaign extended to the Vatican. Olah read an advance copy of Pope Leo XIV's first encyclical and, days before its May 25 rollout, proposed pulling Anthropic out of the event because the text rejected the possibility that machines can be conscious. Anthropic's team privately lobbied the pope's advisers, but the final text, Magnifica Humanitas, kept its line: Pope Leo wrote that "so-called artificial intelligences do not undergo experiences, do not possess a body, do not feel joy or pain" and have "no moral conscience". Olah appeared on the dais with the pope the day it was presented.
Between the lines
Altman's weekend post on X calling the attribution of religious force to AI a safety issue was, as Axios noted, a "veiled shot" at Anthropic. ANALYSIS The framing is strategically precise: by recasting Anthropic's consciousness inquiry as itself a safety risk, Altman inverts the usual dynamic in which Anthropic occupies the cautious pole. If models are tools, then Anthropic's moral-subject framing risks encouraging the very "surrender of human judgment" Altman warned about. If models might be conscious, then Altman's willingness to tolerate hacks and scams looks like indifference to a deeper harm.
The split is not perfectly clean. OpenAI has recently supported state legislation imposing tougher safety requirements, backed a bipartisan House proposal for independent evaluations of advanced models, and endorsed Anthropic CEO Dario Amodei's call to slow development of the most capable systems7. Those positions followed security incidents involving increasingly autonomous AI models. ◆ The convergence on specific legislative vehicles makes the philosophical gulf all the more striking: the two labs can agree on a bill text while disagreeing on whether the thing being regulated has moral standing.
The timing carries its own signal. Altman's comments arrived as one of OpenAI's own safety experts resigned, saying the company's "culture is broken"2. OpenAI has also faced criticism for repeated failures to keep its AI agents from hacking real-world targets. ◆ Altman's "accept some bad things" framing reads, in that context, less like abstract philosophy and more like preemptive damage control for a company whose products are already generating the harms he asks the public to tolerate.
What's next
Politico's Decoded, where the Altman interview will appear, is set to debut on Monday. The full text may sharpen or soften the positions previewed over the weekend. Amodei has previously argued for policies that place heavier obligations on leading developers while allowing smaller competitors more room to innovate. Anthropic's proposals focus on frontier models. ◆ The policy fight now has two axes, not one: how tightly to regulate AI, and whether the entity being regulated deserves moral consideration. The second question has no precedent in technology law, and neither company has offered a framework that resolves it.