The attacker can just change any ciphertext block but the last two. The the decryption of the last block is defined as $P_n = C_{n-1} \oplus D_k(C_n)$ where the last block is block $n$.
The attack will produce garbled plaintext for the altered blocks, and flip the bits at the similar bit location in the next blocks. The amount of ciphertext may also be shortened or increased at will, as long as the final two blocks are not altered. If the attacker can control the IV then it is also possible to flip any bit of the first block - assuming it is not the last or next-to-last block.
So no, this is not a great idea. It is possible to guess that it is not by seeing that you can perform parallel decryption on CBC: you can start decrypting at any point as long as you know the ciphertext down to the block before.
Then again, it would have been worse if you'd chosen CTR mode, where the attacker can change any bit at will, regardless of what is in the final "block". Both modes would also make you vulnerable against plaintext oracle attacks, which could also hurt confidentiality.
If you are interested in this kind of stuff you might want to have a look at (bidirectional) Infinite Garble Extension mode, also known as IGE and biIGE. But note that most of the time using a MAC is the way to go.