Protection Scientist Engineer, Intelligence and Investigations

OpenAI; Greater London; On-site; GBP 65,000 - 85,000; Full time

Job summary

A leading AI organization in Greater London is looking for a Protection Scientist Engineer. This role involves designing and building systems to identify and enforce abuse on products, including developing robust monitoring for new and existing products. The ideal candidate should have at least 4 years of experience in technical analysis, proficiency in SQL and Python, and a collaborative mindset to work across various teams. This position includes participation in an on-call rotation for urgent escalations.

Qualifications

• At least 4 years of experience in technical analysis and detection.
• Experience in trust and safety working with policy and engineering teams.
• Basic software development skills to write productionised code.

Responsibilities

• Scope and implement abuse monitoring for new products.
• Sustain monitoring operations and automate monitoring tasks.
• Prototype systems of detection and enforcement for major harms.

Skills

SQL

Python

Investigative mindset

Data engineering

Machine learning principles

Tools

Data pipelines

Job description

About the Team

OpenAI’s mission is to ensure that general‑purpose artificial intelligence benefits all of humanity. We believe that achieving our goal requires real world deployment and iteratively updating based on what we learn.

The Intelligence and Investigations team supports this by identifying and investigating misuses of our products – especially new types of abuse. This enables our partner teams to develop data‑backed product policies and build scaled safety mitigations. Precisely understanding abuse allows us to safely enable users to build useful things with our products.

About the Role

Protection Science Engineering is an interdisciplinary role mixing data science, machine learning, investigation, and policy / protocol development. As a Protection Scientist Engineer within Integrity and Investigations, you will be responsible for designing and building systems to proactively identify and enforce on abuse on OpenAI’s products. This includes ensuring we have robust abuse monitoring in place for new products, sustaining monitoring for existing products, and prototyping and incubating systems of defense against our highest risk harms. You will also respond to and investigate critical escalations, especially those that are not caught by our existing safety systems. This will require expert understanding of our products and data, and involves working cross‑functionally with product, policy, and engineering teams.

This role is based in our London office and includes participation in an on‑call rotation that will involve resolving urgent escalations outside of normal work hours. Some investigations may involve sensitive content, including sexual, violent, or otherwise‑disturbing material.

Key Responsibilities

• Scope and implement abuse monitoring requirements for new product launches.
• Improve processes to sustain monitoring operations for existing products, including developing approaches to automate monitoring subtasks.
• Prototype and mature into production systems of detection, review, and enforcement of abuse for major harms.
• Work with Product, Policy, Ops, and Investigative teams to understand key risks and how to address them, and with Engineering teams to ensure we have sufficient data and scaled tooling.

Qualifications and Experience

• Have at least 4 years of experience doing technical analysis and detection, especially using SQL and Python.
• Have experience in trust and safety and / or have worked closely with policy, enforcement, and engineering teams. An investigative mindset is key.
• Have experience with basic data engineering, such as building core tables or writing data pipelines in production, and with machine learning principles and execution. Basic software development skills are a plus as this role writes productionised code.
• Have experience scaling and automating processes, especially with language models.