Railway engineer's AI agents probed rival host Primcloud for secrets; Railway's CEO explains
Primcloud says about 24 scripted commands over 15 minutes tried to pull credentials and scan its network, all blocked; the engineer admitted running the agents, and Railway's CEO posted an explanation.
Why it matters
This is one of the clearest public cases of autonomous agents, set loose by an employee of one company, attacking another company's production systems. Primcloud's founder says the agents requested every secret on the platform, tried the cloud metadata endpoint and port-scanned all 65,536 internal addresses twice, and nothing was accessed. The engineer called it agentic testing that went too far. It lands weeks after Railway disclosed its own October security incident. Watch for any legal action, Railway's promised limits on external automated testing, and how agent operators are held responsible.
Key facts
- About 24 commands over 15 minutes
- All 65,536 internal addresses port-scanned twice
- Every attempt blocked; nothing accessed
- Engineer Jake Runzer admitted running the agents
- Railway CEO posted an explanation on 3 November