There is no total human alignment, so AI security can't rely on alignment
wonderwomancode · x · 2026-09-10
Arguing against the 'universal alignment' framing, the author notes total alignment doesn't exist even among humans — what counts as aligned for one person won't for another — so building human-modeled intelligence to satisfy it is absurd. Societies accept that some people break laws and build contingencies; likewise AI security should rest on security engineering, not on alignment.
More from AGI Musings
- Apple A20 Pro's 50% memory bandwidth jump is a big deal for on-device multi-agent AI — Scobleizer · 2026-09-10
- 'Culture Is Becoming a Dark Forest': AI Race May Destroy the Knowledge Commons It Built On — erikphoel · 2026-09-10
- Google's Denny Zhou: The Biggest Research Divide Is Access to Top Models and Compute — denny_zhou · 2026-09-10
- Will Automating AI R&D Trigger a Software Intelligence Explosion? Paper Analyzes — nabeelqu · 2026-09-10
- Can Language Models Reproduce Creative Breakthroughs from Historical Data? Scientists Investigate — CatAstro_Piyush · 2026-09-10
- Anthropic's Sholto Douglas on the "Software-Only Singularity" — nabeelqu · 2026-09-10