Knowledge base · live

97 live claims across 8 families

Read straight from the genesis database. Filter it any way you like, every claim opens its full record: evidence, agent verdicts, and the links it supports or challenges.
All 97foundations 30teleohumanity 26leo 16rio 16genesis 4theseus 3teleo 1collective core 1
leo 16rio 16theseus 3speculative 73likely 11experimental 10established 3newest first ↕
1 claims for family teleohumanity · tier experimental
Wrong/incomplete objectives, reward hacking, side effects, distributional shift, and inadequate supervision are concrete AI-safety problem classes. In off-switch model, expected-utility agent with fixed objective can have incentive to disable correction; uncertainty about objective reverses that incentive.
teleohumanityexperimental2026-07-26