Machine Learning Infrastructure Engineer (Safeguards Research)
Posted Updated
Description
- Anthropic’s Safeguards team builds the systems that detect and mitigate misuse of our AI models, from individual policy violations to sophisticated, coordinated attacks
- A growing part of that work depends on lightweight detection methods trained on model internals, which let us identify harmful behavior cheaply and at scale
- This work feeds directly into Anthropic’s Responsible Scaling Policy commitments
- We’re looking for an engineer to own the infrastructure behind that research
- This is the tooling our researchers rely on to run experiments, train detection methods, and select detections for launch
- It sits between research and production: researchers depend on it for fast iteration, and our detection systems depend on it for reliable, correct results as our models continue to change
- Running machine learning workloads at our scale often requires solving novel systems problems
- You’ll identify those problems and build the abstractions, pipelines, and tooling that keep the research loop fast as requirements shift underneath you
- Strong candidates will have a track record of solving large-scale systems and data problems and will be excited to grow deep machine learning expertise alongside it
- Build and scale the infrastructure and data pipelines behind Safeguards machine learning research
- Own the training, evaluation, and scoring workflows researchers use, with a focus on cutting the time between an idea and a result
- Design tooling and interfaces, including libraries and command line tools, that researchers can use directly without needing to understand the systems underneath
- Build correctness and sanity checking into the stack, so results stay trustworthy as models and workloads evolve
- Take the highest-value research workflows from experiments to reliable, production-grade jobs
- Improve the throughput, cost, and reliability of large-scale inference and scoring workloads
- Partner closely with researchers and engineers across Safeguards to understand their workflows, anticipate how their needs will change, and design for that ahead of time
Benefits
- Comprehensive health, dental, and vision insurance for you and your dependents
- Inclusive fertility benefits via Carrot Fertility
- 22 weeks of paid parental leave
- Flexible paid time off and absence policies
- Mental health support for you and your dependents
- Competitive salary and equity packages
- Optional equity donation matching at a 1:1 ratio, up to 25% of your equity grant
- Retirement plans with competitive matching
- Life and income protection plans
- $500/month flexible wellness and time saver stipend
- Commuter benefits
- Annual education stipend
- Home office stipends
- Relocation support for those moving for Anthropic
- Daily meals and snacks in the office
- Ability to debug performance and correctness problems across an unfamiliar stack
- Experience building tooling or infrastructure that other engineers or researchers use as a dependency
- Strong software engineering fundamentals and hands-on coding ability, with proficiency in Python
- Strong written and verbal communication skills, and a collaborative approach to technical decisions
- Experience building and operating data-intensive or distributed systems in production
- Comfort working across the research-to-deployment pipeline, from exploratory experiments to production systems
- Minimum education: Bachelor’s degree or an equivalent combination of education, training, and/or experience
- Experience building experiment tracking, caching layers, or evaluation harnesses for research teams
- Familiarity with language modeling and transformers, including working with model internals
- Experience with high-performance, large-scale machine learning systems
- Required field of study: A field relevant to the role as demonstrated through coursework, training, or professional experience
- Experience with probes, interpretability, or classifier development
- Experience with machine learning framework internals, GPU or accelerator programming, or inference optimization
- Interest in the misuse risks of AI systems and a desire to work on mitigating them
Skills
Sign in to read and join the discussion.