Field Notes
AI Safety & AGI Risk
This page focuses on AI safety, AGI risk, and the guardrails required to prevent catastrophic outcomes as advanced systems scale.
This isn't hype or doom. It's about understanding where systems break, how incentives fail, and what practical brakes actually look like in the real world.
Why This Exists
Small design or incentive failures scale dangerously at AI levels. What seems like a minor oversight in a narrow system becomes a structural vulnerability when deployed across millions of decisions.
Most catastrophic risks come from deployment and oversight gaps, not malice. The systems we build are often well-intentioned but poorly constrained—and the consequences compound faster than we can course-correct.
This work exists to slow things down before irreversible mistakes happen. Not to halt progress, but to ensure the progress we make is progress we can survive.
Failure Modes
Where alignment, evaluation, and deployment fail quietly at scale.
Guardrails
What should not ship without review, oversight, or hard constraints.
Governance & Incentives
Why speed beats safety — and how policy, norms, and structure can change that.
Podcast
Conversations with builders, researchers, and operators focused on prevention, not prediction.
This page complements the existing Insights section. Longer essays, breakdowns, and evolving thoughts live there, while this page serves as a focused hub for AI safety and AGI risk work.