Although SHA-1 theoretically has collisions, HMAC-SHA-1 which is based on SHA-1 is still widely used (in TLS for example) and is considered to be secure. How is that possible?
As shown in the paper Ricky Demer linked in the comments, HMAC can be secure even when the underlying hash function is not collision resistant. Only PRF-ness of the hash function is required, and SHA-1 is not known to lack it. Or a couple of other conditions can suffice even if it isn't a PRF.
Intuitively, it makes sense that HMAC is secure as a MAC even with SHA-1, because a MAC does not allow a collision search. The only way to find the key would be to compromise the preimage resistance of SHA-1. HMAC in turn prevents length extension attacks and the like that would allow a forgery without knowing the key.
As an aside, even HMAC-MD5 hasn't been broken. However, for the same reason as there – attacks only get better – I would recommend not choosing SHA-1 as the HMAC hash for new applications if you can just as easily use SHA-2.