Tons!
The AT prefix, as such, does nothing whatsoever for error detection or recovery. If we add an AT prefix to an N-character command, the possibility of corruption in the N-character command part is exactly the same as before. Now, the AT also has to be correctly received, too.
An example of a textual protocol whose framing actually means something is the Mobitex MASC: http://www.mobitex.com/resources/masc/MASC_Guide.pdf
MASC frames have a checksum and length field in addition to the framing (frames start with ^ and end with a CR). MASC also dictates the use of 7 bits with even parity; so parity checking is possible by both endpoints. Still, that leading is a fluff feature that contributes nothing; the protocol could work fine with frames just being sequences of non-CR characters terminated by CR.
Modems usually have a solid connection to the host, often a short one; it's the actual end-to-end traffic that is susceptible to corruption; not so much the local commands between the host and modem. These modern LTE modem is often integrated onto the same chip as the host processor which is sending it the AT commands.