OpenAI · posted · checked
Data Scientist, Integrity
PythonSQL$265k – $380kOn-site · San Francisco
Who can apply
On-site · San Francisco
Location: San Francisco
ABOUT THE TEAM
OpenAI’s Integrity team protects people and businesses using ChatGPT, Codex, and the API from fraud, abuse, and harmful misuse. Our work spans payment and promotional abuse, compromised accounts and API credentials, business and advertiser integrity, and emerging misuse of frontier AI capabilities. We also help product teams measure authentic usage so that abuse doesn’t distort decisions about how our products are growing.
Integrity Data Science helps teams decide where to intervene and whether those interventions work. A payment control can prevent fraud while blocking legitimate purchases. A product launch can make it easier for customers to get started while creating new opportunities for abuse. We measure both sides of those decisions.
ABOUT THE ROLE
As a Data Scientist on Integrity, you’ll find emerging abuse patterns, size their impact, and lead the analytical work behind decisions about detection, enforcement, and product launches. You’ll work with Product, Engineering, Research, Policy, and Operations to turn findings into improvements and measure what happens after they ship.
Investigate anomalous activity across products. Estimate how common a problem is, who is affected, and what it costs users or the business. Use that evidence to recommend where to focus.
Define outcome metrics and guardrails, establish baselines, and build monitoring that surfaces changes in abuse, customer friction, and operational performance.
Help build and evaluate detection rules, classifiers, and human review workflows. Use labeled samples and review outcomes to understand what the systems miss, where they flag legitimate activity, and how those errors vary across users and products.
Design randomized holdouts and use quasi-experimental or observational methods when randomization is impractical. Estimate harm or fraud losses prevented alongside effects on legitimate users, revenue, compute costs, and review workload. Recommend whether to launch, expand, adjust, or retire a control.
Trace where signals, review decisions, or enforcement actions break down. Work with partners to close gaps and build reusable datasets, analyses, and investigation tools.
IDEAL CANDIDATE
You use AI tools to investigate large, messy datasets, test hypotheses quickly, and automate recurring analyses. You’re fluent in SQL and Python and can verify AI-generated queries, code, and conclusions against the underlying data.