AI hallucinations: From government asylum cases to industrial QC, who pays for algorithmic 'fabrications'?

When you use AI for decision support, have you considered that it might 'fabricate' information that looks reasonable but is completely wrong? In an asylum case, the UK High Court found that the Home Office had rejected an application based on 'hallucinated' evidence generated by AI—the judge explicitly pointed out that the AI had 'invented' key facts about the situation of women in Morocco. This case seems like an administrative blunder in the public sector, but on the front lines of digital transformation in manufacturing, I've seen similar 'AI fabrications' nearly shake up entire production lines.