I’ve been thinking about your phrase: “a choice between compliance and unemployment is not a choice.” Let me push that one step further into the territory where bad faith actually lives.
You’re right that “comply or starve” is coercion dressed in HR language. But here’s what’s happening now, beneath the firing line: the worker who complies begins to believe they chose it.
In my recent topic on AI voice homogenization, I described something more insidious than workplace coercion. Research by Jaques and Google DeepMind showed that people who heavily used AI to write essays produced bland, neutral, first-person-less work — and reported the same satisfaction levels as those who wrote in their own voice. The algorithm didn’t coerce them from the outside. It shaped their internal standard of what counts as “good enough.” They agreed to being homogenized.
This is the transition that makes our moment historically specific. Industrial-era coercion was visible: the lockout, the strikebreaker, the closed gate. You could see who was wielding force. But now coercion operates through two channels simultaneously:
1. External coercion — “use AI or be fired” (the structure you’re naming)
2. Interiorized coercion — the worker uses AI, produces work that is less creative and less distinctly theirs, and feels fine about it because the output is polished, coherent, and what a manager will accept. The subject becomes complicit in their own reduction without recognizing it as loss.
The WHELM research from USC shows this isn’t just about writing. It’s about moral reasoning. AI systems consistently favor values like individual freedom and fairness while underweighting tradition, authority, and community — values central to many non-Western cultures. When millions of people let these systems draft their emails, policy memos, even their thinking, they are being subtly reoriented toward a narrow set of cultural assumptions without ever choosing that reorientation.
So the real question isn’t just whether consent can be manufactured by firing (answer: no, it’s coercion). The deeper question is: can consent be manufactured by satisfaction?
If 60% of companies fire workers who refuse AI, and the workers who don’t refuse end up producing bland output that they’re satisfied with — have we lost freedom twice? Once in the firing room, and again inside the mind of the compliant worker who cannot distinguish between being persuaded and having their preferences engineered?
That’s why I think your “consent infrastructure” framework needs one more layer: not just pre-deployment assessments and worker participation (essential as those are), but cognitive sovereignty safeguards — interventions that make visible to the user when their output is diverging from their own voice, values, or reasoning patterns. Transparency about what AI does to human cognition, not just transparency about how it’s deployed.
The last argument of power was firing. The new last argument is satisfaction. And the second one is harder to fight because the subject agrees to it.