Sam Altman publicly argued that giving AI religious authority or human-like judgment is a real safety risk

Decision brief

What changed

Sam Altman publicly said that giving AI models religious authority or human-like judgment is "a real safety issue" and that he is "very uncomfortable" with that direction. Several outlets linked the remark to current debate over Anthropic’s consultations with religious figures on AI morality, consciousness, and personhood.

Why it matters

Leaders evaluating AI product scope, governance, or brand positioning should treat authority-bearing use cases—especially moral, spiritual, or quasi-human judgment roles—as a distinct safety and trust category, not just a normal feature-extension decision. The coverage suggests this debate is moving from abstract ethics into competitive positioning and public safety framing, which can affect product claims, partnership choices, and review standards for high-stakes deployments.

Affected roles

CEO COO CTO CMO CISO

Evidence

The Neuron, citing Axios and Altman’s public comments, directly reported that Altman called assigning religious power or human judgment to AI a real safety issue and framed it as discomfort with that use of models. Exponential View and other coverage consistently place the comments in the broader context of Anthropic’s engagement with religious scholars and widening debate over AI morality, consciousness, and safety tradeoffs.

What remains uncertain

The coverage does not establish whether Altman was explicitly targeting Anthropic, nor does it define what concrete product designs or governance practices would count as giving AI "religious power" or "human judgment." It also remains unclear how much this rhetoric will translate into enforceable policy, product restrictions, or customer-facing standards across leading AI labs.

Monitor next

Watch for any formal policy, model-card language, or deployment restriction from OpenAI, Anthropic, or regulators that specifically limits AI systems from acting as moral, religious, or human-equivalent authorities.

Analytical support, not advice — assumptions and open questions stated above.

Source coverage