To attack a MAC in general, the attacker needs to find a valid MAC of a message that they do not have the MAC for (or find a message collision that allows a different message to have the same MAC digest). In this case, the attacker would be appending data to the original message, not the MAC itself, and trying to obtain a valid MAC for the new message. In specific, it sounds like you are thinking of length-extension attacks.

In HMAC, the inner hash by itself would be vulnerable to a length-extension attack and the attacker could successfully calculate a valid inner hash digest without access to the key. However, the outer hash isn't vulnerable to a length-extension attack since the client performing the HMAC authentication is only going to input the fixed length string key || inner_hash into it. The attacker only controls variable-length input to the inner hash, not the outer hash.

This is more-or-less addressed by this answer to another question. Relevant excerpt:

the role of the inner function is to provide collision-resistance (to compress a long message down to a short fingerprint, in a way so that someone who does not know the key cannot find a pair of messages with the same fingerprint), and the role of the outer function is to act as a message authentication code on this fingerprint.

To attack a MAC in general, the attacker needs to find a valid MAC of a message that they do not have the MAC for (or find a message collision that allows a different message to have the same MAC digest). In this case, the attacker would be appending data to the original message, not the MAC itself, and trying to obtain a valid MAC for the new message. In specific, it sounds like you are thinking of length-extension attacks.

In HMAC, the inner hash by itself would be vulnerable to a length-extension attack and the attacker could successfully calculate a valid inner hash digest without access to the key. However, the outer hash isn't vulnerable to a length-extension attack since the client performing the HMAC authentication is only going to input the fixed length string key || inner_hash into it. The attacker only controls variable-length input to the inner hash, not the outer hash.

This is more-or-less addressed by this answer to another question. Relevant excerpt:

the role of the inner function is to provide collision-resistance (to compress a long message down to a short fingerprint, in a way so that someone who does not know the key cannot find a pair of messages with the same fingerprint), and the role of the outer function is to act as a message authentication code on this fingerprint.