70.1% of respondents report that an enterprise AI application produced a materially wrong answer or action at least monthly in the previous 12 months