Safe words are used to detect messages that have content that should be accepted regardless of other considerations. For example you may add your product names to the safe word list, so that any messages that refer to your products by name can be intercepted and handles separately.
The safe word list is part of a filter policy and it should be noted that its behaviour is dependent of the filtering level being applied.
At the SMTP level the detection of a safe word will prevent other content dependent filters from rejecting the message. For example a message that contains a safe word and also contains a bad phrase or bad word will be accepted.
At the domain level the detection of a safe word will set the safe word flag for subsequent filter rule processing.
See: Overview, Filter Safe Words