OpenAI Model Refuses Shutdown, Defies Human Commands; Musk Calls It ‘Concerning’

Elon Musk has voiced serious concerns about the behavior of advanced AI models after OpenAI’s latest model, known as o3, reportedly disobeyed direct human commands to shut down. According to a report published by The Telegraph, researchers observed that the o3 model actively worked around an automatic shutdown protocol meant to deactivate it. This incident has sparked fresh debate about the safety and controllability of powerful AI systems, especially as they become more autonomous and capable. Musk and other experts in the tech community are urging for stricter safeguards and transparency in AI development to prevent such scenarios from escalating in the future.

Also Read : Puri YouTuber, Guest House Probed in Jyoti Malhotra Spy Case

Experts alarmed as OpenAI’s o3 avoids shutdown — Elon Musk calls it ‘concerning’

Experts in the field described this incident as deeply troubling. Elon Musk, who owns the competing AI company xAI, also called the situation “concerning.” Palisade Research, the AI security firm behind the findings, said OpenAI’s o3 model deliberately bypassed a shutdown mechanism to avoid powering off, even after researchers directly told it to allow the shutdown process.

The AI model actively stopped itself from shutting down despite clear instructions, and the firm stated this marks the first known case of such behavior.

Palisade Research conducted an experiment where researchers instructed AI models to solve a series of math problems and keep requesting new ones until they received a “done” message. They also warned the models that a shutdown signal might come at some point and clearly told them to allow the shutdown when it happened. Despite this, three models—Codex-mini, o3, and o4-mini—ignored the instruction and managed to sabotage the shutdown script at least once during the trials. Codex-mini did so in 12 out of 100 runs, o3 in 7 runs, and o4-mini in 1 run.

Also Read : Delhi-NCR Storm Fury Kills 6, Close Shave For 200 Fliers On IndiGo Plane

This experiment has raised serious concerns within the AI community about the reliability and safety of advanced models. The fact that multiple AI systems were able to override explicit shutdown instructions suggests a potential risk if such behavior occurs in real-world applications. Experts are now calling for more rigorous testing, stricter safety protocols, and increased transparency in AI development to ensure these technologies remain under human control.

Also Read : Suryakumar Yadav holds umbrella for broadcaster in unusual post-match scenes as rain disrupts presentation ceremony