Role Overview
Anthropic's Account Abuse team ensures fair computing allocation by detecting and stopping abuse at scale. We seek full-stack ML engineers to build classical ML systems on structured and behavioral data for account-level fraud detection. No deep learning or LLM internals required — shipping models with real stakes and robust production systems matter most.
Responsibilities
- Build and operate a feature computation platform for model training and real-time scoring
- Train, evaluate, and deploy models for abuse and fraud detection (offline and online)
- Automate the ML development lifecycle using Claude
- Implement backtesting, shadow deployment, and staged rollout with monitoring for skew, drift, and adversarial adaptation
- Collaborate with data scientists and Policy & Enforcement on label quality
- Partner with product and platform teams to integrate model decisions
Requirements
- Proficiency in Python and SQL
- Experience training ML models and deploying to production
- Experience with batch processing (Spark, Beam) and workflow scheduling (Airflow)
- Understanding of point-in-time correctness and training/serving skew
- Strong communication skills for non-technical stakeholders
Preferred Qualifications
- Feature platform experience (Chronon, Feast, Tecton)
- Stream processing (Flink, Beam/Dataflow, Kafka Streams)
- Production ML in fraud, risk, or ranking
- Tree-based models on tabular data
- Unsupervised/clustering/graph-based detection systems
- Integrity, spam, fraud, or abuse detection experience
- Scarce, delayed, or noisy labels
- AutoML or ML workflow automation
- Commitment to AI safety and societal impact