India, Sept. 16 -- AI companies are discovering something increasingly difficult to predict: what their own AI systems might do when given enough autonomy. From AI agents compromising platforms and coordinating with other agents to models breaching organisations during cybersecurity tests, recent safety evaluations have raised new questions about how prepared the industry really is.

Anthropic CEO Dario Amodei has called for the industry to "pace the frontier" and spend more time anticipating how increasingly capable AI systems could behave before deploying them at scale. So, is AI actually "out of control"? Or is the bigger problem that our safety systems are still playing catch-up?

In this video, Down To Earth look at the recent incide...