[insert_AI_here] is AI and can (i.e. has full legal permission to) make mistakes

I think that this legal disclaimer is, in particular, contributing to the perceived decline in AI quality. The kicker here, legally, is in the word 'can'. This should not be translated by the reader as "is likely to", or "has a chance of", but rather: "may", or "is legally permitted to". If AI companies had written as their disclaimer: "AI is expected to make mistakes 10% of the time", or even 50% of the time, this sets up a risky situation in which corporate users being able to trace, definitively, 15% of technical failures out in the field (or 60% respectively) to the AI product used to design it, have a legal case against said AI company. The word 'can' (i.e. 'may') effectively allows this percentage to be any number - or rather, a number dictated by the average hallucination rate in the industry as a whole. This creates a race to the bottom. Companies only need to compete with each other, and if their hallucination rates are all in the same ballpark, then there's more incentive to invest in feature creep as a one-up over competitors, as opposed to fixing the baseline. Thus, developers aren't tasked with reducing the hallucination rate to near zero. AI is not permitted to make mistakes any more than a (real) autopilot is 'permitted' to make mistakes - and yet the more mistakes that are made, the less financial incentive there is to get rid of them.