Traditional single-user security models do not necessarily capture the power of real-world attackers. A scheme that is secure in the single-user setting may not be as secure in the multi-user setting. Inspired by the recent analysis of Schnorr signatures in the multi-user setting, we analyse Boneh-Lynn-Shacham (BLS) signatures and Boneh-Gentry-Lynn-Shacham (BGLS) aggregate signatures in the multi-user setting. We obtain a tight reduction from the security of key-prefixed BLS in the multi-user model to normal BLS in the single-user model. We introduce a multi-user security model for general aggregate signature schemes, in contrast to the original “chosen-key” security model of BGLS that is analogous to the single-user setting of a signature scheme. We obtain a tight reduction from the security of multi-user key-prefixed BGLS to the security of multi-user key-prefixed BLS. Finally, we apply a technique of Katz and Wang to present a tight security reduction from a variant of multi-user key-prefixed BGLS to the computational co-Diffie-Hellman (co-CDH) problem. All of our results for BLS and BGLS use type III pairings.