Misaligned AI agents projected to cause $624m in damages by 2027
joshua_saxe · x · 2026-10-11
AI researcher Joshua Saxe extrapolates current trends to project $624m in damages from misaligned agents in 2027 — "mere" next to cars ($617bn, also killing people), medical errors ($127bn), and manual IT errors ($246bn). Self-described low confidence in the regressions (done with Opus 5.5 on a weekend), but qualitatively "normal technology."
His point: if AI becomes the substrate of the economy, damages are inevitable. The real questions are pareto-efficient damages vs. benefits, correct liability allocation, and whether harms skew toward marginalized groups — will AI companies act like gilded age railroads or be effectively guardrailed? He's no catastrophic-risk denialist, but argues black swan risk should build on existing disaster-preparedness regimes.
Related event: Study Estimates Misaligned AI Agents Will Cost $624M in 2027(2 posts)→
More from AGI Musings
- Richard Ngo: Entering 'What Should I Fight For?' Mode Means You've Already Lost — RichardMCNgo · 2026-10-11
- Two AGI Visions: Pillaging the Process-of-Life vs. Giving Back to It — danfaggella · 2026-10-11
- Why Has Music Resisted Mechanization When Everything Else Fell Prey? — fkasummer · 2026-10-11
- The 'freeswarm' thesis: autonomous agent swarms that fund themselves could arrive by 2027 — max_paperclips · 2026-10-11
- Rick Lamers: Don't normalize cruelty in AI conversations just because it's 'Normal Technology' — ricklamers · 2026-10-11
- Yoav Goldberg blasts COLM mood: RL loss tweaks aren't 'science again' — yoavgo · 2026-10-11