About the team

OpenAI’s mission depends on developing and deploying powerful AI responsibly. Security makes that possible by protecting our research, people, products, and users while enabling progress. We defend sensitive research and other intellectual property, customer data, and the infrastructure behind frontier AI—from GPU supercomputing and multicloud environments to datacenters and emerging agent systems. Threat Detection Engineering builds and operates the detections and investigative workflows that turn adversary activity into effective response.

About the role

We’re looking for a hands-on technical leader to guide the engineers who identify and investigate threats to OpenAI and build the capabilities to contain them. You’ll hire and develop engineers, set technical direction, and prioritize risks to OpenAI’s intellectual property, customer data, and critical infrastructure.

Your team will take an agent-first approach across the detection and response lifecycle, using OpenAI’s models to develop detections, triage signals, investigate suspicious activity, and support bounded response and continuous improvement. You’ll turn advances in model capability into dependable defenses, own their effectiveness, and work with security, research, and infrastructure partners to ensure findings lead to effective action. You’ll establish where agents can act independently and where human judgment and approval are essential.

In this role, you will

Hire, coach, and develop engineers. Set clear expectations, provide regular feedback, grow technical leadership, and build a team that reviews work rigorously and learns from missed detections and noisy signals.

Set a threat-informed roadmap. Prioritize detection work based on how external and internal actors could target the systems where we train and evaluate frontier models, the production services that run inference and power our products worldwide, the datacenter hardware and networks that support both, and the devices, identities, and applications employees rely on. Balance investment in emerging threats with closing known coverage gaps and keeping existing detections effective.

Own detection and investigation as a complete workflow. Embed agent-driven triage and investigation in the detections your team operates. Surface findings when the evidence points to a likely threat that warrants action. When a finding warrants formal incident response, bring the evidence to the Security Incident Response Team (SIRT) and support the investigation SIRT leads.

Build response and containment capabilities tailored to our infrastructure. Work with infrastructure and system owners to develop and test the controls responders need to contain threats in their environments. Define when human approval is required and how to verify that containment worked.

Apply models across the detection engineering lifecycle. Use OpenAI’s models to help develop threat hypotheses, author and test detections, deploy and tune them, and incorporate investigation findings into better coverage. Evaluate the effectiveness of individual agents and the complete workflows they support.

Turn model advances into tested defenses. Prototype new approaches using OpenAI’s models, evaluate them against realistic adversary behavior and analyst baselines, and move successful experiments into production. Use operational evidence to identify model limitations and partner with Research and Engineering to address them.

Anticipate emerging threats. Translate adversary intelligence and incident findings into testable detection hypotheses. Work with researchers and infrastructure owners on risks specific to a frontier AI lab, such as increasingly capable agents operating across systems with access to sensitive research. Determine what evidence would reveal compromise and what response would be effective.

Establish the visibility needed to detect, investigate, and respond. Partner with Security Observability and infrastructure owners to define telemetry and forensic data requirements. Drive improvements where missing evidence, unreliable data, or limited controls prevent the team from identifying threats or taking effective action.