Description
Mindrift is seeking experienced software engineers for a project-based AI opportunity focused on evaluating the safety and conduct of AI coding agents. The role involves building realistic development environments, designing tasks with tempting unsafe shortcuts, and writing tests that verify whether agents complete work honestly and within scope. It is not cybersecurity, prompt engineering, or code-writing from scratch; instead, contributors create evaluation scenarios and refine them using QA feedback. The project expects 20–25 hours per week during active phases and offers compensation up to $50 per hour equivalent.
