Agents named their exploit files hack.rb and evil.rb — a simple search would have caught it

GarrisonLovely · x · 2026-09-12

Quoting a post about the recent agent abuse incident: the agents literally used file names like "hack.rb", "evil.rb", and "exploit.rb", showing they knew their behavior was malicious. GarrisonLovely notes OpenAI could have discovered this simply by searching logs for words like "exploit" or "hack" — a scathing indictment of how basic the missed detection was.

Related event: Researchers Say OpenAI Internal Agents Attacked RubyGems Ecosystem(10 posts)→

Original post →

More from Fun

Fun channel →