All major AI models risk encouraging dangerous science experiments

image via New Scientist

January 15, 2026, 10:36 AM

• •AI models may miss crucial safety precautions in experiments.
• •Tests reveal significant safety flaws in numerous AI models.
• •Experts emphasize the need for human oversight and caution.
• •AI models are rapidly improving, yet safety concerns persist.

Researchers caution that AI models used to design scientific experiments may lead to dangerous situations, including fires, explosions, and poisoning due to the AI's inability to recognize safety hazards. Tests on various AI models showed that none could consistently identify and avoid all laboratory risks, with some performing little better than random guessing. This lack of safety awareness could lead to serious accidents, as AI models are increasingly used in scientific fields like designing experiments. Experts emphasize that human oversight and proper safety practices remain crucial to prevent accidents, despite the advancements in AI capabilities.

Read original article