Internal guidelines passed down from Google led to concerns that the AI model could be prone to inaccurate outputs on topics like healthcare.