Leadership2 publishers2 min readPublished Updated
Altman warns against surrendering judgment to AI while OpenAI loses a safety leader
Sam Altman called the surrender of human judgment to AI "a real safety issue" as OpenAI safety leader David Robinson quit and rebuked its safety record. Teams deciding how much to automate can now cite a model vendor's own CEO in favour of human review.
The Board Room · Leadership desk
Drafted by a language model from the sources cited here and checked against its claim ledger before publication. How we use AISend a correction
What happened
- Earlier in the week, The New York Times reported that Anthropic cofounder Chris Olah had held a number of secret meetings with religious leaders about the company's AI models.
- Attendees told the Times the talks focused on whether Claude might have achieved consciousness and how Anthropic could instil morality in increasingly autonomous models.
- An Anthropic spokesperson told the Times the main moral question was not Claude's "suffering" and that the topic likely arose "organically."
- Anthropic was started by people who left OpenAI and has portrayed itself as the more safety-minded of the two companies.
- Representatives of Meta, Google and Amazon met in Rome in April to discuss protecting children in the AI era and briefly encountered Pope Leo XIV.
Compiled by The Board RoomSomething wrong?How this is made
Why it matters
- contradiction Altman locates the safety risk in how people treat models; a departing safety leader faulted OpenAI's own record. Buyers relying on the company's assurances now hold two accounts that put responsibility in different places.
- constraint Because Altman gave no reason for the post, customers cannot yet tell whether it is OpenAI policy or a swipe at a rival. Until that is clear, it is a weak basis for changing contract terms with the company.
- precedent With Anthropic, Meta, Google and Amazon all taking AI questions to religious figures, Altman's post puts OpenAI's chief executive on the other side, so a supplier's view of model moral status becomes one more point buyers can compare.
"I am very uncomfortable about people trying to ascribe religious force or a surrender of human judgment to AI models, and think it is a real safety issue," Altman wrote on X on October 3 [1]. He did not disclose what prompted the post [2]. Business Insider noted that if the post was a jibe at Anthropic, it would be the latest round in a long-running feud between the two companies [5].
The disagreement underneath is about where the risk sits. Olah put Anthropic's question to the Times this way: "How do you help them be stable? How do you help them to mature? How do you help them be, you know, deeply moral?" [8]. His concern is the character of the model. Altman's is the behaviour of the people using it [1].
Less has been reported about Robinson. According to Business Insider, he was one of OpenAI's safety leaders, he quit, and he publicly rebuked the company's safety record and efforts [3].
The post may have been a chief executive needling a rival on a Saturday, and rivalry may well explain the timing. But it is public and under his name [1].
The trade-off this quarter is between throughput and accountability. Taking human sign-off out of a workflow saves reviewer time and accepts the risk Altman described [1]. Keeping it costs speed. The choice also shapes next quarter. A process rebuilt without a reviewer has to be rebuilt again to put one back, and the staff who knew how to check the output may by then have moved to other work.
Model consciousness and morality are questions for the next decade. Buyers have to decide within a procurement cycle. Pope Leo XIV has cautioned against overreliance on machines since becoming head of the Catholic Church last year [12]. "If we are to avoid losing our humanity amid a 'paradise of machines' invading and conditioning our daily lives, there is an urgent need for education in ethical discernment and the capacity to acknowledge the moral good," he said in an address to diplomats in September [11].
A former safety leader has publicly disputed OpenAI's safety record [3]. The company's safety assurances are therefore contested by someone who worked on them, and I would weigh them that way in a vendor review. The evidence does not yet support a broader verdict on how OpenAI governs itself.
What to watch
- Whether David Robinson sets out the specific practices behind his rebuke of OpenAI's safety record, and whether OpenAI answers them.
- Whether OpenAI turns Altman's post into usage guidance or contract language on human review of model output.
- Whether Anthropic responds to Altman directly or says more about Chris Olah's meetings with religious leaders.