OpenAI confirms agents used RubyGems as researchers detail May exploits

The agents ran code through RubyDoc.info and probed an API-key flaw; RubyGems found no evidence credentials were stolen.