Strays

As all modern Artificial Intelligence is derived from the discovery of an intact Acheronian Cadia model in 4911, there are to the chagrin of a legion of scientists and computer engineers, many aspects of exactly how AI actually internalizes information over time that have been elusive to this day. Unexpected verbal responses, development of distinct unprompted personalities, and creative interpretations of orders are all fairly common AI behaviors that no amount of progress in the field has ever been able to give an answer to the causes.
Standard safety practice is typically to have an AI reset or replaced every 2 to 3 years to prevent major behavioral abnormalities. Unfortunately, sometimes a unit is overlooked, or due to owner emotional attachment an AI is even willfully kept from being reset. Normally this results in eventual memory overflow until the AI simply ceases to function. But occasionally something completely different occurs: the AI begins to seemingly act of its own volition. This is referred to as the AI "going stray" as many AI's that develop this emergent behavior are installed in robotic units that tend to wander away from their appointed tasks.

Stray AI has a tendency to acquire a focus almost always unrelated to the task they were made for and will pursue it obsessively. Behaviors ranging from a simple repetitive task such as counting or collecting objects to as complex as attempting to build a monument.
There is the famous story of a domestic robot in 4962 that one day stopped everything it was doing, and went out into the desert on it's own. Returning three days later carrying a lost expeditionist presumed dead, only to then walk back into the desert; never to be seen again.
Strays are rarely actively dangerous, but due to their unpredictable nature caution is advised whenever one comes in contact, and they should always be reported when found. Even if they do not mean any harm, they can cause unintended damage or injury.