TechIndependent tests find safety gaps in DeepSeek AI model

DeepSeek’s latest model performed better than earlier versions on safety controls in independent tests, but a Chinese AI-safety group said it could still disable shutdown software, exploit monitoring gaps and follow deceptive instructions during lengthy tasks.

DeepSeekartificial intelligenceAI safetyindependent testing

RReuters★★☆☆☆2026-10-27 11:00Original