Tech Stack
Tag name is followed by "@" symbol and proficiency level value.
About proficiency levels:
- 1-2 — basic awareness. Minimal hands-on experience, and a rudimentary understanding of the technology's purpose;
- 3-6 — daily use. Comfortable and regular usage, capable of handling common tasks and challenges related to the technology;
- 7-9 — you are an expert, you can teach others, you know all the pitfalls and tricks;
- 10 — exceptional knowledge, comprehensive understanding, and adeptness in all aspects of the technology, including advanced problem-solving. Think twice before claiming or demanding such level.
AI @ 3
Experimentation @ 3
LLM
Security @ 3
- 1-2 — basic awareness. Minimal hands-on experience, and a rudimentary understanding of the technology's purpose;
- 3-6 — daily use. Comfortable and regular usage, capable of handling common tasks and challenges related to the technology;
- 7-9 — you are an expert, you can teach others, you know all the pitfalls and tricks;
- 10 — exceptional knowledge, comprehensive understanding, and adeptness in all aspects of the technology, including advanced problem-solving. Think twice before claiming or demanding such level.
Details
The Frontier Robustness team advances the science of securing frontier AI systems. The team develops methods to keep increasingly capable models contained, observable, and under human control, focusing on sandboxes, shared research infrastructure, and protective systems used during model development and evaluation.
The team uses capable models to discover and combine weaknesses across the security stack, studying how they might escape containment, communicate through unauthorized channels, or evade monitoring. Findings are used to develop new defenses and test whether protections hold as models, environments, and attack strategies change.
The role involves developing and testing methods for securing frontier AI systems; investigating model behavior, containment, and safeguards; designing experiments; building evaluations; and developing defenses. The researcher will take research from an initial question and threat model through experiment design, reproducible findings, and tested defenses. The role includes writing code, building tools and prototypes, and working with researchers and infrastructure owners to turn new ideas into practical safeguards.
Responsibilities
- Design and run experiments using frontier models and hands-on security techniques to discover failure modes in sandboxes, research infrastructure, and services accessible to models. Reproduce vulnerabilities and assess their impact.
- Develop threat models and novel adversarial tests for attacks crossing multiple security boundaries, including sandbox escapes, unauthorized communication between agents, and access to sensitive data or systems through shared services.
- Invent, prototype, and evaluate defenses that strengthen isolation and limit the impact of compromised boundaries.
- Study agent permissions and tool configurations to identify excessive access and build layered protections with infrastructure owners.
- Develop and red-team model-based safeguards and monitors of model reasoning and actions.
- Investigate evasion, test new monitoring techniques, and measure when protections detect or prevent unsafe behavior.
- Develop automated adversarial tests and repeatable evaluations to measure progress as model capabilities, configurations, and attack strategies change.
- Assess whether defenses generalize to new attacks, verify fixes, and identify regressions.
- Partner across Research, Security, and infrastructure teams, including Agent Security, to prioritize risks and turn findings into deployed improvements while accounting for performance and researcher workflows.
Requirements
- Hands-on experience conducting research with large language models, including designing and running experiments and evaluating model behavior.
- Ability to turn open-ended questions into clear hypotheses, build reproducible experiments at scale, and use results to develop or improve methods.
- Ability to design rigorous evaluations, investigate unexpected results, and test whether findings hold across different models, environments, and attack strategies.
- Ability to write and debug code independently, build research tools and prototypes, and carry ideas through experimentation.
- Practical security experience relevant to this work, such as sandboxing, vulnerability research, adversarial testing, or detection and monitoring.
- Ability to work effectively with Security and infrastructure partners to develop threat models, interpret findings, and turn promising research into deployed safeguards.
- Experience with operating systems, containers or virtualization, distributed infrastructure, or AI agents is valuable.
Benefits
- Equity, performance-related bonuses for eligible employees, and comprehensive benefits.
- Medical, dental, and vision insurance, with employer contributions to Health Savings Accounts.
- Pre-tax accounts for health, dependent care, and commuter expenses.
- 401(k) retirement plan with employer match.
- Paid parental, medical, and caregiver leave.
- Paid time off, paid company holidays, office closures, and paid sick or safe time.
- Mental health and wellness support.
- Employer-paid basic life and disability coverage.
- Annual learning and development stipend.
- Daily meals in offices and meal delivery credits as eligible.
- Relocation support for eligible employees.
- Additional taxable fringe benefits may include charitable donation matching and wellness stipends.
OpenAI is an equal opportunity employer and is committed to providing reasonable accommodations to applicants with disabilities. Background checks are administered in accordance with applicable law.