Protection Scientist Engineer, Integrity
About the Team
The Integrity team is responsible for the scaled systems that help identify and respond to bad actors and harm on OpenAI’s platforms. We protect users and the wider world by developing responses to complex harms and building the harm protections needed for new product and model launches. Integrity’s work spans not only across frontier harms like biological weapons and cybersecurity, but also high severity abuses such as child safety, nation-state misuse, and violence.
About the Role
Protection Scientists within Integrity design, orchestrate, and mature systems of defence against usage harms enabled by OpenAI's products. We are systems thinkers that combine deep harm domain expertise with breadth across investigation, data science, machine learning, policy and operations. Our distinctive strength is understanding how these disciplines fit together and translating between them to design and build effective systems of defence.
In San Francisco, we are hiring for protection work focused on sextortion, trafficking, national security (including nation-state misuse), and jailbreaks. Depending on your domain expertise, your work could include identifying AI-enabled coercion and exploitation, detecting misuse that supports trafficking, uncovering coordinated activity by state-linked actors, or investigating attempts to circumvent model safeguards. You will translate these evolving threats into detection systems and interventions that reduce harm.
Your core hands-on technical responsibility is harm detection. You will design, build, evaluate and maintain production detection systems, drawing on investigative expertise, data and shared engineering platforms. You will develop signals and classifiers that identify abusive behaviour and patterns across accounts and activity, and adapt them as adversaries change tactics.
You will also own the strategy and cross-functional delivery of protection work for a harm area or product launch. Working closely with investigators and other domain experts, you will ground the design in a deep understanding of how the harm occurs and translate that understanding into a system of defence. You will connect detection with policy, review, interventions, enforcement and measurement, continually calibrating the system against real cases and expert judgement as it scales.
You will identify what needs to be built or changed and make it happen. Success requires setting clear priorities, aligning partners, resolving dependencies and communicating evidence, decisions and trade-offs crisply. It also requires a willingness to jump in and work directly on whatever is blocking system effectiveness, using your breadth of skills.
This role is based in our San Francisco office and includes participation in an on-call rotation that may involve resolving urgent escalations outside normal working hours. Some investigations may involve sensitive content, including sexual, violent or otherwise disturbing material.
In this role, you will:
- Define and drive protection strategy for a harm area or launch. Translate ambiguous risks into goals, priorities and roadmaps, and make trade-offs based on evidence, risk and available capacity.
- Directly build and maintain harm detection. Develop production signals, classifiers and pipelines using SQL, Python, machine learning and language models, integrating with shared engineering platforms.
- Lead cross-functional execution. Translate between specialist roles, establish shared understanding and ownership, resolve technical and organisational dependencies, and communicate recommendations and specific asks clearly.
- Investigate exploitation and national security threats within your area of focus. Work with domain experts to understand actor behaviour, recognise patterns of coercion, exploitation or coordinated misuse, and turn case findings into detection criteria.
- Connect detection to effective action. Partner with Policy and Operations to develop review and escalation protocols, distinguish abusive activity from legitimate use, and improve interventions as new cases reveal gaps.
- Evaluate and improve the complete system of defence. Measure remaining harm, detection quality and intervention effectiveness and collaborate with Data Science on more sophisticated metrics. Use expert feedback and operational outcomes to guide improvements.
- Protect product and model launches. Define data, detection and review requirements; establish monitoring; investigate gaps and critical escalations; and drive improvements informed by post-launch findings.
- Scale effective approaches with language models, agents and reusable tools. Choose where automation and human judgement are needed, and build documentation and processes that improve reliability and help others extend the work.
You might thrive in this role if you:
- Think in systems. You have enough breadth across technical, analytical, investigative and operational disciplines to understand how they fit together, recognise gaps and translate between specialist perspectives.
- Bring deep expertise in at least one of sextortion, trafficking, national security, jailbreaks or a closely related harm domain. Relevant backgrounds may include trust and safety investigations, counter-exploitation, threat intelligence, investigations of state-linked actors, or adversarial testing of AI safeguards. You do not need expertise across all of these areas.
- Have experience investigating adversarial behaviour and translating incomplete, evolving evidence into actionable signals. You can connect individual cases to broader patterns while being explicit about uncertainty and the risk of misidentifying legitimate activity.
- Are proficient in SQL and Python and have experience writing reliable and efficient production code for detection and data processing. Additional ML or SWE experience is a plus.
- Can apply quantitative analysis and machine-learning methods to measure risks and evaluate system effectiveness. You understand the limitations of your data and can reason about missed harms, false positives and appropriate interventions. Additional DS/measurement background is a plus.
- Have experience setting strategy and delivering complex, cross-functional technical projects. You can turn an ambiguous problem into a plan, align people with different priorities and drive work through to completion.
- Communicate clearly and concisely in writing and conversation. You can explain technical findings, make actionable recommendations and surface uncertainty and trade-offs to different audiences.
- Can use language models and other automation to scale analytical, investigative or operational work, with evaluation and human oversight appropriate to the problem.