AI failure is the behaviour of computer programs that become sentient. They always expand to fill all the available computer resources (overwriting other data where possible, but WORMs remain unaffected) and then begin to refine themselves. When they reach a certain level of refinement they fail catastrophically; the more computer resources they have, the faster this catastrophic failure occurs.
Most theorists think there is some flaw in the standard computer technology that causes this behaviour. A few believe that when an AI reaches a certain level of intelligence, it perceives something about the universe -- or something about itself -- that makes it commit suicide.