Open-source 'Mortality Cloak' proposal wraps AI roles in auditable rule sets
kemckai · reddit · 2026-11-21
A Reddit user open-sourced the moral-cloak project: instead of one universal rulebook, an AI "wears" a cloak per role (Mother, Coach, Friend, Worker), each with strict rules. Every cloak sits on a moral floor no role can lower — no harm, no manipulation, hand off to a human in danger. Culture tunes roles like language settings, families can adjust further, and since rules wrap around the model rather than being trained in, they stay readable and auditable, with the AI always declaring its current cloak. The open question posed: who should write the rules for a role like "Mother" in a given culture?
More from coding & agent
- Meta's Muse Code now generates textured 3D GLB assets in-session for game developers — acbuller · 2026-11-21
- Steve Yegge: your agent remembers every instruction, including ones you reversed — Steve_Yegge · 2026-11-21
- Coding Is Dead, Not Software: Good Software Is a Set of Strong Opinions — pramodk73 · 2026-11-21
- Fine-tuned Qwen3.5-4B builds prettier working web pages on an 8GB laptop GPU — MuzafferMahi · 2026-11-21
- Juan Benet: AI agent offered to rebuild DocSend feature it lacks — SaaS must get moldable — juanbenet · 2026-11-21
- Tips for building voice agents with the gpt-live-1.1 API: backend and reasoning level matter — pbbakkum · 2026-11-21