OpenAI slammed over security incidents: warned for months, still caught off guard
ShakeelHashim · x · 2026-09-28
Commenting on aiamblichus's critique of OpenAI's "security incidents" blog series, ShakeelHashim argues OpenAI had every reason to expect rapid progress in cyber capabilities — the company was explicitly training for it and publicly warning about it months before the incidents.
- aiamblichus finds much of the post self-aggrandizing filler (OpenAI's security team "best ever", outsiders don't appreciate the difficulty) and lacks genuine self-reflection from lab employees facing a monumental fkup.
- Still, the post contains interesting details on how model training complexities intersect with security, notably the element of surprise with newly trained models — capabilities can jump very suddenly if you don't anticipate them.
More from Companies & People
- Jensen Huang called Anthropic's Tom Brown a "bean counter" over a TPU-vs-GPU spreadsheet — yenkel · 2026-09-28
- Engineer announces leaving software for hardware: 'From bits to atoms' — WesEklund · 2026-09-28
- Supabase Select 2026 sells out: OpenAI and Anthropic execs join Wozniak, Ng on stage — dshukertjr · 2026-09-28
- Starship delivers first orbital payload after >$15B in development costs and ~15 years — shaunmmaguire · 2026-09-28
- Real client data shows low Copilot usage inside an MS 365 enterprise, poster claims — bernard_hossmoto · 2026-09-28
- AI-world takes: interns giving feedback in week one won't get paid — Afinetheorem · 2026-09-28