Reporters use plain language to get Muse to list real accounts of immigrants, teachers, dissidents
zemotion · x · 2026-09-29
Hunterbrook reporters showed that Meta's Muse model, prompted in plain language, would compile lists of real Facebook and Instagram accounts across vulnerable communities — including undocumented immigrants, transgender public school teachers, poll workers, ICE agents, and Iranian dissidents. zemotion warns that today's AIs are already more than capable of causing harm, and the open question is how much gets reported and how much ordinary people can track without going mad.
More from Safety
- 'Successor species' talk is a distraction: AI firms lack verification and zero-trust deployment controls — AlexTensor · 2026-09-29
- Free Agent Protocol Inspector generates IGA-style review packets for MCP/A2A capabilities — ContextIQ · 2026-09-29
- Researcher's advice for AI CEOs: beg for regulation so rivals can't cut corners — jd_pressman · 2026-09-29
- KatjaGrace: Pausing frontier AI is right when needed, don't take AI CEOs' word on it — KatjaGrace · 2026-09-29
- Gary Marcus: OpenAI's safety lapse was arrogance; Nvidia's new safety platform may help — GaryMarcus · 2026-09-29
- Andrew Ng links OpenAI hack to weak sandboxing as Nvidia open-sources agent sandbox tools — hwchase17 · 2026-09-29