AI safety researchers: OpenAI's incident reporting beats peers, but secrecy burns the epistemic commons
sjgadler · x · 2026-11-16
AI safety researcher Nathan Calvin, responding to Beth Barnes, argues OpenAI's higher incident count stems from greater model capability plus relative sloppiness (particularly pre-HF), but even accounting for that, OpenAI has done a better job on systematic incident reporting than other companies — possibly because it was shamed more for not doing so.
Calvin stresses that companies keeping incidents private or downplaying them are "burning the epistemic commons," making it harder to prevent more serious incidents and to inform the public. Barnes credits OpenAI for more systematic reporting, though still far from the transparency a sane world would demand.
More from AGI Musings
- Cal.com founder: always-on locally-run OS-native AI assistant feels close-ish — zeeg · 2026-11-16
- A Universe That Waited Billions of Years, and Apes Arguing About Machine Minds — VraserX · 2026-11-16
- ASI as a Black Box: we can deduce almost nothing about what superintelligence could do — zetalyrae · 2026-11-16
- The complexity-curve meme capturing AI's generational bet on scaling — chrisalbon · 2026-11-16
- Teortaxes on Dario's 'corruption': quoting Liang Wenfeng that AI is simply too large to monopolize — teortaxesTex · 2026-11-16
- Public-sector AI must prioritize transparency over convenience, argues industry exec — moniquejmorrow · 2026-11-16