You can't use current spam filters to fight against this, provided the system is implemented correctly.
Let's assume you send 1000 automated messages periodically, and they contain a mixed bag of normal looking (so that your genuine comms could hide in plain sight) and malicious messages (to make their detection systems sweat).
All the messages are sent from the same origin (e.g.: a messaging app capable of this), with a specific interval (the genuine msg is timed accordingly), etc..
Let's see a small sample of normal looking messages, with an added genuine manually written message:
- Let's meet up at the Riverside Park soccer field at 5 PM sharp.
- Hey, can you grab a gallon of milk from the corner store on your way back?
- Good morning! Wishing you a fantastic day ahead, especially during your 11 AM meeting!
- Just a reminder: our dinner reservation at Luigi's Italian Bistro is at 7 PM tonight.
- Could you pick up the kids from Maplewood Elementary at 3:30 PM today? Work is keeping me busy.
- Caught in traffic near Main Street, but I'll be at the coffee shop in about 10 minutes!
Which one is the genuine one?
Just like you cannot discern which msg is genuine (are any of them?), you can generate plausible maliciously looking messages as well.