Meta's Muse AI stuff was not only obviously a bad idea, its "fleet learning" stuff was extra especially an obviously bad idea. But look *how* bad it was: https://www.reuters.com/investigations/meta-bets-big-muse-ai-agent-early-safety-mishaps-surface-2026-11-18/
- Muse sets up doctor's appointments for other peoples' medical conditions with their email addresses and names
- Muse, without prompting, giving money laundering advice
- Muse telling you about other peoples' bank account status
- Muse being more than a little creepy about kids you don't have
EDIT: One caveat in that it's not totally clear whether or not these are hallucinations or a crossover of data between accounts. Regardless, the level of confusion here is alarming given the amount of credentials over peoples' lives users are encouraged to hand over.