"Right 95% of the time but can't prove when it's right" is the best statement of the adoption ceiling I've seen.
One extension is say... visible reasoning traces build trust in the session, but the sanctions scenario happens months later, in front of a judge, outside the vendor's UI. At that point the question isn't whether the lawyer could see the sources... it's whether the firm can produce an independent record of what was retrieved and relied on.
That's the difference between a trust feature and an evidence chain.
"Right 95% of the time but can't prove when it's right" is the best statement of the adoption ceiling I've seen.
One extension is say... visible reasoning traces build trust in the session, but the sanctions scenario happens months later, in front of a judge, outside the vendor's UI. At that point the question isn't whether the lawyer could see the sources... it's whether the firm can produce an independent record of what was retrieved and relied on.
That's the difference between a trust feature and an evidence chain.