I'm not an expert by any means - you'd need Joe (my supervisor) for that! - but here's a relatively common example that he pointed at:
https://www.acnlabs.co.uk/blog/neuro-bollocks-alert-dopamine..."Muscle movements on the head (like moving your eyes, smiling, gurning, etc) produce electrical noise that is far stronger than the brain signals EEG tries to measure. You cannot put a helmet over an EEG kit and then bang your head around in a jet or race car and get anything but pure noise."
That's a crude example, but it's the kind of over-interpretation that's really common with these things.
And that's using a proper EEG skullcap, not these ridiculously simple things that you slip over your head without any conduction gel etc. In my non-expert opinion a large dataset of this kind of vague data would just lead to mostly mush because the signals that come from these things are so rubbish. But it won't stop people who want to make money from it that there are clear signals there, like the use of Myers-Briggs, astrology etc.
The only current technology that actively concerns me is direct neural implantation. You can get really clean, clear signals from that depending on where you hook into. Anything else feels like it's a long way away from anything which would actively harm privacy, although clearly people have a vested interest in making that work and we shouldn't be complacent.