Security Researcher, Agentic AI Threats

As AI agents become more capable at software engineering, and automate more of our internal work, they could become a dangerous cyber threat. People in this role will help OpenAI prepare for security threats from advanced AI agent insiders.

Responsibilities

Identify paths by which capable future internal AI agents could compromise OpenAI; Design security controls - focusing on measures with long lead times that benefit from advanced preparation; Stress-test defenses with AI agent evaluations and penetration tests

Requirements

Deeply technical across security and modern infrastructure, and comfortable digging into the details of operating systems, cloud, containers, CI/CD, or distributed systems; Strong software engineering skills and enjoy building prototypes yourself; Interested in engaging with stakeholders and can do so effectively

Nice to have

experience securing cloud infrastructure, and are deeply familiar with core components of the AI stack