When somebody says that something might be true, it’s intuitive to think of that claim as licensed by that person’s own information: when you say something might be true, you do so because it might be true as far as you know. This intuitive interpretation is generally held to be rendered untenable by disagreement facts like the following:

A: Paul might be at the party.

B: You’re wrong, he’s in Barbados.

If A’s assertion makes a claim about A’s own knowledge—that Paul being at the party is compatible with what A knows—B’s rejection of that claim should mean that Paul being at the party is not compatible with what A knows. But instead, B can reject A’s claim due to Paul being at the party not being compatible with what B knows. So, the argument goes, A’s claim cannot be a claim about A’s knowledge, despite the intuition that A gets to assert it on the basis of her own knowledge. I show that this argument doesn’t actually work. The problem that disagreements like this pose for a speaker-oriented semantics for might is a technical artifact of how we’ve chosen to formalize our theory of assertive updates. I show that an alternative formalization of our theory of assertive updates dissolves the problem, suggesting that these sorts of disagreements pose a challenge for the theory of assertion, not for the semantics of epistemic modals.