In most Bantu languages, an object prefix can occur on the verb. In some Bantu languages, this object prefix has a purely anaphoric function, while in others it has an additional agree- ment function. Since Bresnan & Mchombo, Chichewa (Bantu N.31 Malawi) has been con- sidered a textbook example of a language where the object marker is “always an incor- porated pronoun and never a non-referential marker of grammatical agreement” (Bresnan & Mchombo 1987: 755). That is, in order for an overt nominal phrase (DP) to co-occur in the same sentence with an object prefix, the DP must be a dislocated Topic. Conversely, a dislocated object DP (a Topic) must be anaphorically bound to an object prefix. In this pa- per I present new Chichewa data showing that in modern colloquial Chichewa there is a human/non-human asymmetry in object marking. Human object DPs commonly co-occur with an object prefix, whether the object is a dislocated Topic or not, whereas non-human ones commonly do not co-occur with an object prefix, even when they are dislocated Top- ics. I conclude that Chichewa shows differential object marking (or object indexation), as hu- manness is a more important condition on the occurrence of object prefixes than word order. The implications of the Chichewa (and other Bantu) data for recent proposals like Creissels (2006), Dalrymple & Nikolaeva (2011) and Iemmolo (2013; 2014) about the diachronic devel- opment of DOM agreement systems from anaphoric Topic marking systems are discussed, and an alternative constraints-based account is proposed.