Yes, that is precisely the danger it represents, and the reason its presence is a red flag in an investigation.
If you are hooked up to the (older) machines, nervous, and it starts swinging wildly when you are asked a shocking question, any reasonably bright person understands the machine is indicating a physiological reaction that is associated with deception. But when they say they "failed", they are talking about "failing" to control autonominous nervous system functions, because, and it can't be said too often, correlation is not causation*.
Further, the unexamined underlying philosophy behind it's deployment is huge problem. In polygraph-land, crimes aren't puzzles or riddles to be solved, but competitions where the investigator "wins" when everyone perceives them to have "caught the bad guy." While correlation is still not causation, polygraphs appear more than chance would indicate in stories about coerced false confessions. Coerced false confessions are the inevitable consequences of Investigation as Completion.
And polygraphs are a leading indicator of Investigation as Competiton, and hence worse than no evidence at all.
(*This kind of amplified mis-understanding is the situation with the nothing-burger Dedmon text messages- they are not "confessions", they are genuinely innocent people, the girls at any rate, who don't understand they have been targeted by people willing to go to any lengths to convince the public of someone's guilt, true or not, as opposed to determining the truth of their guilt.
Investigation as Competiton.)