Research Scientist Intern, Integrity (Trust and Safety) AI (PhD)

Meta was built to help people connect and share, and over the last decade our tools have played a critical part in changing how people around the world communicate with one another. With over a billion people using the service and more than fifty offices around the globe, a career at Meta offers countless ways to make an impact in a fast growing organization. Meta is seeking Research Interns to join the Integrity (Trust and Safety) team within Facebook. We are the applied ML team developing cutting-edge AI solutions to detect harm from bad actors. Our models decide when a group spreading severe harm should be taken down, when a creator is faking engagement, when a page is impersonating someone, and when an enforcement should be reversed on appeal, across a platform used by billions of people. Our work has moved from per-post classifiers to advanced agentic AI systems that reason over an entire entity (its content, members, admins and history) and apply written policy the way an expert would, pushing the precision/recall boundaries on adversarial actor detection. The open research problems are about making that judgment efficient, reliable: -Pushing the performance boundaries for agentic AI applied to bad actor detection -Distilling large reasoning models into solutions that can scale cost-efficiently, including pushing the boundaries of decision-style models like Jev -Improving shared multi-harm model improving without regressing other harms Interns own one focused research problem end to end, from framing to experiments to publication, under the guidance of our research scientists and machine learning engineers. You will own E2E ML workflows including AI-driven agentic workflows with Gemini/Claude/Muse as well as scaled solutions through data analysis, feature engineering, model development and deployment. Responsibilities Perform research to advance the science and technology of trustworthy, efficient AI systems for integrity and safety. Develop novel methods for training, distilling and continually updating large language and multimodal models that make policy decisions about entities and actors. Design rigorous experiments and evaluations using expert human labels, LLM judges and offline replays, including methods that stay valid when labels are scarce or expensive. Analyze and improve the efficiency, calibration, robustness and reliability of models used in production enforcement. Collaborate with researchers, engineers and policy partners, communicating research plans, progress and results. Publish research results and contribute to research that can be applied to Meta product development. Minimum Qualifications Currently has or is in the process of obtaining a Ph.D. degree in Computer Science, Artificial Intelligence, Machine Learning, Natural Language Processing, Computer Vision, Statistics, or relevant technical field Must obtain work authorization in country of employment at the time of hire and maintain ongoing work authorization during employment Experience with Python Experience with deep learning frameworks such as PyTorch Research experience in machine learning, including training or fine-tuning deep learning models and designing experiments to evaluate them Familiarity with large language models and AI-native development (Claude, Codex etc) Preferred Qualifications Intent to return to the degree program after the completion of the internship/co-op Proven track record of research achievements as demonstrated by publications at leading venues such as NeurIPS, ICML, ICLR, ACL, EMNLP, NAACL, COLM, CoLLAs, KDD, WWW or ICWSM Research or project experience in one or more of: continual or lifelong learning, multi-task learning and model merging, active learning, knowledge distillation, LLM post-training (SFT, preference optimization, RL), uncertainty estimation and calibration, learning from AI feedback or LLM judges, multimodal modeling, or sequence/graph modeling of user behavior Experience applying ML to trust and safety, content moderation, fraud or abuse detection Experience with large-scale or parameter-efficient training (e.g., LoRA, distributed training) Demonstrated software development experience via internships, work experience, coding competitions, or widely used open-source contributions Experience working and communicating cross-functionally in a fast-paced environment, with the ability to frame real-world problems as rigorous ML problems and drive iterations from clear analysis $7,650/month to $12,667/month + benefits