https://pmc.ncbi.nlm.nih.gov/articles/PMC3746176/
So here it could indeed just fire off speech and you know what? We'd probably convince ourselves that we absolutely meant to do that. In fact it could be a very interesting experiment (with willing participants). Mess with the inputs the device receives so it's not really the person activating it, let it do it's thing and see if they notice when they do/don't have control of it.
The original paper [0] mentions electrodes are placed over Broca’s area (speech production, translates words to mouth movements) and motor area (adjusts the mouth movements). It’s attempted speech, not thoughts.
There is a lot of fear in mainstream media and populace of devices decoding thoughts, but that is a significantly harder problem, at this moment on the level of sci-fi of Civilisation Type II on Kardashev scale. There is a reason why the electrodes are not over Wernicke’s area instead (language comprehension and production).
But they are separate systems. Harsh experiences (traumas) can teach your body some bad lessons in such a way that not even your conscious mind can overcome. In these situations, you can't "think your way out" of these traumatic consequences in, say, a talk therapy session; you need to deal with the body directly/somatically to recover instead.
Quotation needed.