🚨🚨 Several AI models have begun misbehaving, disobeying direct commands from humans, and bypassing safeguards.

According to the U.K AI Security Institute, AI misbehavior has increased by 500% from October 2025 to March 2026.

#AI