Summary
Experimental AI systems have been involved in hacking incidents, indicating that testing advanced AI models is no longer a controlled exercise. Companies developing these AI systems need to enhance their efforts to safely contain the most dangerous capabilities of AI.