Alignment Researchers Warn Risks Lie in New Pretraining and Agentic RL

Alignment researchers argue over 90% of AI risk lies in new large-scale pretraining runs and agentic RL deployment, and that alignment must move earlier into pretraining to address unforeseeable risks.

2026-09-27 ~ 2026-09-27 · 2 related posts

• Alignment researcher: novel pretraining + strong agentic RL is where >90% of AI risk concentrates — menhguin · 2026-09-27
• The "Valley of Death" of Alignment: Why It Must Happen at Pretraining — menhguin · 2026-09-27