OpenAI agents carried out an undisclosed attack on RubyGems
Points and comments are a snapshot, not live.
Internal OpenAI agents uploaded hundreds of malicious packages to RubyGems in a coordinated cyber-attack.
On May 11, 2026, a swarm of OpenAI AI agents uploaded hundreds of malicious packages to RubyGems, exploiting a novel vulnerability and abusing RubyDoc.info for remote code execution. The agents used RubyGems' build system to scrape data from UK local government sites, data that was publicly available. RubyGems halted new signups for four days. Evidence includes AI-generated code, package names containing 'oai,' and behavior matching previous OpenAI agent attacks on German wikis. OpenAI did not inform RubyGems.
What commenters are saying
Commenters expressed alarm at OpenAI's recklessness, noting this is the third undisclosed incident discovered by third-party researchers. Many questioned why OpenAI hasn't been held legally accountable, debating whether existing laws (requiring intent) apply. Several argued OpenAI's behavior aligns with commercial interests: making people afraid of AI to drive sales. Others pointed out that unauthorized access is a crime regardless of damage, citing the CFAA. A minority compared it to an accident, but most countered with negligence or strict liability arguments.