The evidence supports this as a claim about current AI, not as a timeless rule. Existing AI systems have no established way to detect or confirm their own consciousness, and their self-descriptions are better explained as generated outputs than privileged self-knowledge. The statement overreaches only because some philosophical accounts leave open the possibility that a different, genuinely conscious AI could know this in principle.
Caveats
The wording is too broad: it is well-supported for current AI systems, but not proven as a universal truth about all possible AI.
An AI’s verbal claim that it is conscious is not strong evidence of self-knowledge; current systems can produce such statements through pattern generation.
Some philosophical views allow that a future AI with the right architecture could, in principle, have genuine self-knowledge of consciousness.