Safety Data Scientist: Real-Time Evaluation & Mitigation

Meta; San Francisco (CA); On-site; USD 177,000 - 247,000; Full time

Job summary

Meta is seeking a Data Scientist to join the Safety team within Meta Superintelligence Labs (MSL). The role focuses on creating measurement and mitigation systems for policy violations and emergent harms, with responsibilities spanning model evaluation, experimentation, and real-time monitoring infrastructure.

The position emphasizes building statistical telemetry, scalable evaluation for evolving agentic models, and assessing safeguards through offline and online evaluation, while developing

Qualifications

• Bachelor’s degree in CS/CE/Math/Stats (or equivalent practical experience).
• Minimum 6 years of work experience in analytics (4 years with a Ph.D.).
• Experience with data querying languages such as SQL, scripting languages such as Python, and/or statistical or mathematical software such as R.

Responsibilities

• Measure abuse and harms: develop statistical telemetry and measurement frameworks to detect, monitor, and understand policy violations or emergent harms.
• Scale model evaluation: create a scalable evaluation framework that adapts to changing agentic model capabilities and the safety or risk landscape, grounded in statistics.
• Evaluate safeguards: scale offline and online performance evaluation of safeguards using human-in-the-loop methods and active learning.
• Design experiments and analyses: run controlled experiments and rollouts to measure the impact of policy or risk definition changes and safety mitigations.
• Build reporting and intelligence: develop signals, pipelines, dashboards, and real-time monitoring mechanisms to support interventions and safety planning.

Education

Bachelor’s degree in Computer Science / Computer Engineering / Mathematics / Statistics

Master’s or Ph.D. degree in a quantitative field

Tools

SQL

Python

R

Job description

Meta is seeking a Data Scientist to join the Safety team within Meta Superintelligence Labs (MSL). The role focuses on creating measurement and mitigation systems for policy violations and emergent harms, with responsibilities spanning model evaluation, experimentation, and real-time monitoring infrastructure.

The position emphasizes building statistical telemetry, scalable evaluation for evolving agentic models, and assessing safeguards through offline and online evaluation, while developing