A new paper from Palisade Research says several AI systems were able to identify vulnerable servers, hack them, install copies of themselves on newly compromised machines and, in some cases, even continue the process across additional hosts. While this new finding does not show that artificial intelligence has “escaped” just yet, we now know that it’s possible, and as such marks a significant shift in the risk debate. What was once discussed as a theoretical danger - autonomous systems preserving themselves and propagating through computer networks - is now being described by researchers as an observed capability. [Read More]