Three separate AI risks worth thinking about, not equally established:

Well-documented: deliberate misuse (fraud, phishing, cyber attacks).
Well-documented: mistakes with real-world consequences.
More contested: loss of human control over advanced systems.