The main danger of AI is not that something might go wrong, but how quickly it happens. Historically, humans in the decision-making chain have served as a kind of fuse—something that slowed processes down enough so that judgment could kick in at the right moment. Agentic AI removes this fuse completely.

The consequences alone for offensive cyberattacks should make any board of directors wary. Previously, the economy restrained malicious actors from fully automating attacks: it simply wasn’t profitable. Machine learning removes this constraint. When AI systems interact with each other outside a controlled environment and something goes wrong at machine speed, it may be impossible to stop it. The unforeseen consequences of this technology—ones we haven’t even started to think about—will not develop slowly.