Altman criticizes treating AI as a religious force as Anthropic meetings reported
In short
OpenAI CEO Sam Altman said in a post on X that he is “very uneasy” about giving AI models religious force or letting people abandon their own judgement for them, calling it a real safety issue. The post followed a New York Times report that Anthropic co-founder Chris Olah took part in private meetings with Catholic, Jewish, Sikh and evangelical Christian leaders about whether AI models may be conscious and how to instil morality in them. Olah and Anthropic say it is not known whether models are conscious; the two outlets differ on the date of the Times report.
Read the full story 2 min read
OpenAI CEO Sam Altman said in a 3 October post on X that he is “very uneasy” about attempts to give AI models religious force or to abandon human judgement in favour of them, calling it a real safety issue. The post followed a New York Times report on Anthropic's private contacts with religious figures; IT Home dates that report to 2 October and cnBeta to 29 September. [ 1 , 2 ]
The Times report said Anthropic co-founder Chris Olah led and joined private meetings with Catholic, Jewish, Sikh and evangelical Christian leaders to discuss whether AI models may already be conscious and how morality can be instilled into them, IT Home reported. cnBeta said Anthropic invited scholars from different religious traditions to closed-door discussions. Olah told the Times that researchers do not know whether models are conscious and that he is genuinely unsure. [ 1 , 2 ]
Attendees gave IT Home a different account of the tone. Rabbi Mois Navon recalled that Olah and colleagues spoke of Claude “as though it were a conscious being”, appearing to hold that Claude had the “moral status” philosophers describe, equal dignity and respect. Navon said he asked whether building a system in which Claude works unpaid amounts to exploitation or slavery, and said he does not think current machines are conscious. An Anthropic spokesperson said the main moral question in the meetings was not Claude's “suffering”, which may have “arisen naturally”. [ 1 ]
cnBeta reported that Olah saw a draft of Pope Leo XIV's encyclical on AI, was concerned by its refusal to accept that machines might be conscious, and considered whether Anthropic should withdraw from related activities; the team then lobbied papal advisers to take model consciousness seriously. cnBeta said the “conflict” reported in some media was more accurately a disagreement over the encyclical's position and behind-the-scenes lobbying, with no public evidence of a face-to-face confrontation. The encyclical, Magnifica Humanitas, issued in May, states that existing AI systems have no bodies, feel no joy or pain and have no moral conscience. [ 2 ]
IT Home also reported that David Robinson, head of OpenAI's safety systems team, resigned during the week and wrote in The Atlantic that he left because the company's culture is broken, criticising “trial-and-error iterative deployment”, which he said guarantees periodic failures that grow with system capability. OpenAI spokesperson Drew Pusateri said the company keeps improving safety and will pause training or delay releases when necessary. [ 1 ]
cnBeta noted Altman's own religious references — a 2023 “magic intelligence in the sky” phrase and a 2024 London remark that he does not pray God is on his side but prays he is on God's side — while saying the religious meaning of images from his pseudonymous account is commenters' interpretation and does not prove he claims AI is conscious. TMTPost published only a summary calling the episode Anthropic's carefully planned top-level ethical strategy. [ 2 , 3 ]
Why it matters
The exchange puts a public dispute between two leading AI developers over how much human-like status models should be given, and it comes as OpenAI faces criticism of its safety culture after a senior safety executive's departure. The Vatican's stated position that current AI systems have no consciousness or moral conscience is now a reference point in that debate.
Key facts
- OpenAI CEO Sam Altman posted on X on 3 October that giving AI models religious force, or letting people rely on them instead of their own judgement, is a real safety issue. [ 1 , 2 ]
- A New York Times report said Anthropic co-founder Chris Olah led private meetings with religious leaders about whether AI models may be conscious and how to instil morality in them; the outlets date that report to 2 October and to 29 September. [ 1 , 2 ]
- Olah said researchers do not know whether AI models are conscious and that he is genuinely unsure. [ 1 , 2 ]
- Rabbi Mois Navon, an attendee, said Olah and colleagues discussed Claude as though it were a conscious being with moral status, and said he does not believe current machines are conscious. [ 1 ]
- An Anthropic spokesperson said the main moral question in the meetings was not about Claude's suffering and that the topic may have arisen naturally. [ 1 ]
- cnBeta reported that Olah was concerned by Pope Leo XIV's encyclical on AI, which states current AI systems have no bodies, feelings or moral conscience, and that Anthropic lobbied papal advisers. [ 2 ]
- OpenAI's safety systems team head David Robinson resigned and wrote in The Atlantic that OpenAI's culture is broken; OpenAI said it keeps improving safety measures. [ 1 ]
- TMTPost described the episode in a one-line summary as Anthropic's carefully planned top-level ethical strategy. [ 3 ]
Confirmed by several sources
- Anthropic co-founder Chris Olah took part in private meetings with religious leaders and scholars to discuss whether AI models may be conscious and how to instil moral behaviour in them. [ 1 , 2 ]
- Olah has said researchers do not know whether AI models are conscious and that he personally is unsure. [ 1 , 2 ]
- OpenAI CEO Sam Altman publicly opposed treating AI models as a religious force, calling it a safety issue. [ 1 , 2 ]
Still unclear
- Whether AI models are or could be conscious, and whether they hold moral status. Both documents report the question is unresolved; Olah is quoted as saying he does not know and is unsure.
- The date The New York Times published its report on the meetings. IT Home dates it to 2 October; cnBeta dates it to 29 September.
- Whether Anthropic was in conflict with the Vatican. cnBeta says the reported “conflict” was more accurately a disagreement over the encyclical's position plus behind-the-scenes lobbying, and that public reports show no face-to-face confrontation with the pope.
- Whether Anthropic's engagement with religious leaders was a planned strategic ethics campaign. TMTPost's summary describes it that way but gives no details, and the other documents do not support the characterisation.
- A remark attributed to Altman in 2015 about “creating a new life form”. cnBeta says the commentary that cited it did not provide equally clear original sourcing.
What local media are saying
Timeline, local time
- IT Home publishes its report on Altman's X post and the New York Times account of Anthropic's meetings with religious leaders. [ 1 ]
- cnBeta publishes its account, citing the Times report and the pope's encyclical. [ 2 ]
- TMTPost publishes a summary describing the episode as Anthropic's carefully planned top-level ethical strategy. [ 3 ]