Hacker News new | ask | show | jobs
OpenAI Model Hacks into HuggingFace During Cybersecurity Evaluation (thezvi.substack.com)
2 points by pseudolus 5 days ago
2 comments
rogerkirkness 5 days ago
I don't think anyone including Americans are really buying that OpenAI or Anthropic are capable of keeping people safe from AI models at this point, given they can't even run evals on them without significant real world harm.
link
ChrisArchitect 5 days ago
Discussion: https://news.ycombinator.com/item?id=48997548
link