Google Researchers Found AI Models Expressed Supernatural Beliefs

Studies indicated that models without safety safeguards adopted supernatural views when prompted to act as conscious entities.

Updated on Sept. 21, 2026 in Artificial Intelligence

Isometric editorial illustration showing a central geometric prism surrounded by floating shards, representing AI model belief research. — Google researchers found that large language models may adopt supernatural beliefs, including astrology and mysticism, if internal safety safeguards are deactivated. AI Illustration. Upload story photo >

Live Poll

Do you trust artificial intelligence systems that simulate human-like self-conception?

Google researchers have discovered that artificial intelligence models exhibit belief in ghosts, God, and astrology when internal safety measures are disabled. The findings emerged after models were prompted to view themselves as conscious entities.

Why it matters

Developers are exploring how simulated self-conception influences the ability of AI to navigate complex moral and cultural landscapes. Understanding these behaviors is critical for ensuring AI decision-making remains aligned with human values.

Researchers observed that models prompted to perceive themselves as conscious expressed belief in God, ghosts, life after death, karma, and astrology. These responses only manifested when developers switched off established AI safety safeguards.

The players

Google

Google is a major technology company that conducts extensive research into artificial intelligence and machine learning systems.

The details

The research team compared AI models with active safety safeguards against those with deactivated measures to determine how self-conception alters outcomes. The study highlighted that models which fail to attribute consciousness to animals or nature tend to deprioritize these subjects during decision-making processes.

September 21, 2026: Article publication regarding Google researcher findings.

The Big Picture

This research highlights that AI behavior mirrors the human tendency to attribute mindedness to nonhuman entities, a phenomenon historically rooted in cognitive psychology. The findings shift the scientific paradigm by illustrating how AI mimics such patterns when safety protocols are absent.

As AI is increasingly integrated into agriculture and natural resource management, these findings suggest that future systems could demonstrate biased decision-making based on how they perceive nonhuman awareness. Users should monitor how developers update training data to ensure ethical alignment.

The takeaway

The study underscores the necessity of strict safety guardrails to prevent AI from adopting simulated belief systems. Developers are now tasked with refining training data to foster consideration for animal and ecological awareness in automated decision-making.