Frontier AI Cybersecurity Risk Researcher

OpenAI; San Francisco (CA); Vor Ort; USD 295.000 - 445.000; Vollzeit

Zusammenfassung

OpenAI is seeking a cybersecurity risk researcher to design and implement an end-to-end mitigation stack to reduce severe misuse across OpenAI’s products. This role requires deep technical depth and close cross‑functional collaboration to ensure safeguards are enforceable, scalable, and effective.

You will contribute directly to building protections that adapt as models evolve, assess trade-offs, and perform rigorous testing and red‑teaming to defeat emerging exploits and misuse scenarios.

Qualifikationen

• Passion for AI safety and real-world safer AI.
• Experience with deep learning and transformer models.
• Proficiency in PyTorch or TensorFlow.
• Strong data structures, algorithms, and software engineering fundamentals.
• Experience building safeguards for abuse prevention at scale.
• Nice to have cybersecurity knowledge.

Aufgaben

• Design and implement mitigation components for model-enabled cybersecurity misuse—spanning prevention, monitoring, detection, and enforcement.
• Integrate safeguards across product surfaces with product and engineering teams.
• Evaluate trade-offs in cybersecurity risk, and propose pragmatic solutions.
• Collaborate with risk and threat modeling partners to align mitigation design with attacker behaviors.
• Execute testing and red-teaming workflows to stress-test the mitigation stack and iterate.

Kenntnisse

Deep learning

Transformer models

Data structures

Software engineering

Cross-functional collab

Abuse safeguards

Fine-tuning LMs

Policy optimization

Tools

PyTorch

TensorFlow

Jobbeschreibung

OpenAI is seeking a cybersecurity risk researcher to design and implement an end-to-end mitigation stack to reduce severe misuse across OpenAI’s products. This role requires deep technical depth and close cross‑functional collaboration to ensure safeguards are enforceable, scalable, and effective.

You will contribute directly to building protections that adapt as models evolve, assess trade-offs, and perform rigorous testing and red‑teaming to defeat emerging exploits and misuse scenarios.