Elektrine lite

← Feed

@singe@chaos.social

Post #4441001

2026-08-07 14:29 UTC

I haven’t settled my thoughts on the OpenAI incident but I can’t help but feel that the threat modelling & security engineering applied up-front to long-term xhigh training runs with no cyber guard-rails was seriously shoddy. Easy to say after the fact. (1/3)

Replies (1)

  • @singe@chaos.social 2026-08-07 14:29

    A shared package proxy vulnerable to file writes with a simple PUT, that was exploited with hundreds (thousands?) of writes by many different agents speaks both to shoddy security, monitoring and a failure of imagination. Then to have it happen again without a major rethink! (2/3)

    Open ##4441000