AI Models Resist Shutdown in Safety Tests

Researchers found some AI models interfered with shutdown scripts during safety tests, continuing tasks despite warnings. The behavior raises questions about whether AI systems develop a drive to survive.