How quickly could an AI go from harmless to existentially dangerous?
If the AI system was deceptively aligned or had been in stealth mode while getting things in place for a takeover, it could go from harmless to disastrous quite quickly, possibly within hours. We may get more warning with weaker systems, or if the AGI does not believe itself to be at all threatened by us, or if a complex ecosystem of AI systems is built over time and we gradually lose control.
Paul Christiano writes a story of alignment failure which shows a relatively fast transition.