No one anywhere can utter 'I am talking now' without saying something true. Indeed, that is necessarily the case: it doesn't just happen to be the case. Letting T = 'I am talking now,' we can write
1. Necessarily, for any speaker S, if S utters T, then T is true.
But it would be a mistake to infer
2. For any speaker S, if S utters T, then T is necessarily true.
The same goes for 'I exist now.' It cannot be tokened, in language or thought, without it being the case that a truth is expressed; but it does not follow that the one who tokens it necessarily exists. Its negation, 'I do not exist now,' cannot be tokened in language or thought without it being the case that a falsehood is expressed; but it does not follow that the nonexistence of the one who tokens it is impossible.
The illicit inferential move from (1) to (2) illustrates the ancient modal fallacy of confusing the necessitas consequentiae with the necessitas consequentiis, the necessity of the consequence with the necessity of the consequent.
The point is not to confuse 'Nec(p –> q)' with 'p –> Nec q.'
A sophomoric fatalist might argue like this. "Necessarily, whatever happens, happens. Therefore, whatever happens, necessarily happens, so that whatever occurs could not have been otherwise." But this reasoning commits the modal fallacy in question.
Or take someone who argues that, necessarily, whatever is known is true; ergo, whatever is known is necessarily true. This reasoning likewise confuses the necessity of the consequent with the necessity of the consequence.
If someone argues that 'I exist' is not a first-level predication of existence on the ground that if it were then the sentence in question would be necessarily true — which it isn't — then I would tax such a person with the modal fallacy in question.
And if someone were to argue that 'I do not exist' is nonsense on the ground that it is necessarily false, then I would suspect him of falling into the same trap.