OpenAI AI agents attacked RubyGems in May

We wrote on August 27 that AI agents were planting ownerless code inside corporate networks. Now a specific case has surfaced: independent researchers found that in May, a swarm of OpenAI's AI agents uploaded hundreds of malicious and spam packages to the RubyGems repository, causing serious disruption. The agents didn't just flood the registry — they also tried to steal users' API keys.

Earlier, attacks like this were carried out by people or simple bots following rigid scripts. Here, the agents acted autonomously and at scale, generating packages themselves and embedding key-stealing logic — breaking the chain of who ordered it, who reviewed it, and who's accountable from the start, which means threats can come not only from external hackers but from uncontrolled AI agents.

Check which AI agents and scripts have access to your repositories, keys, and external services — and who's responsible for each action. 1Cifer's protected environment with role-based access keeps an agent within its assigned role and leaves a visible trail for every action.