Security Labs Engineer

About Anthropic Anthropic’s mission is to create reliable, interpretable, and steerable AI systems. We want AI to be safe and beneficial for our users and for society as a whole. Our team is a quickly growing group of committed researchers, engineers, policy experts, and business leaders working together to build beneficial AI systems. About the Role Security at Anthropic is not a compliance exercise. It is a core part of how we stay safe as we build increasingly capable systems. Our Responsible Scaling Policy commits us to launching structured security R&D projects: ambitious, time-boxed experiments designed to resolve high-uncertainty questions about our long-term security posture. Each project runs for roughly 6 months with defined exit criteria. Some will succeed and move toward production. Others will fail, and we'll treat that as useful signals. The questions these projects are designed to answer include: Can our core research workflows survive extreme isolation? Can we get cryptographic guarantees where we currently rely on trust? Can AI become our most effective security control? As a Security Labs Engineer, you own one or more projects end-to-end: scoping the experiment, building the infrastructure, coordinating across teams, running the pilot, documenting results, and where the experiment succeeds, helping scale it into production. This is 0-to-1 and 1-to-10 work. Current Project Areas The portfolio evolves based on what we learn. Current areas include: Designing and operating a mock high-assurance research environment: simulating what our infrastructure would look like under extreme isolation and physical security controls, with real measurement of productivity impact Exploring cryptographic verification of model integrity using techniques like zero-knowledge proofs to provide mathematical guarantees about what is running in production Assessing the feasibility of confidential computing across the full model lifecycle (note: this is an open question, not a committed roadmap item) Piloting AI-assisted security tooling including vulnerability discovery, automated patching, anomaly detection, and adaptive behavioral monitoring Prototyping API-only access regimes where even internal research workflows never touch raw model weights Part of your job is helping shape what comes next based on gaps uncovered in the current round. What You'll Do: Own the end-to-end execution of a Security Labs project: refine the hypothesis, design the experiment, build the prototype, run the pilot, and write up the results Build novel security infrastructure under real time pressure: isolated clusters, hardened access controls, cryptographic verification layers, with a bias toward learning fast Where experiments succeed, drive them toward production scale. An experiment that works on one cluster but not a hundred is not a finished result. Work embedded with research teams (Pretraining, RL, Inference) to stress-test whether their core workflows can function under extreme security controls, and document precisely where they break Evaluate and integrate emerging security technologies through coordination with external vendors and research groups Turn experimental results into clear, decision-ready writeups that inform Anthropic's long-term security architecture and RSP commitments Maintain a pain-point registry and feasibility assessment for each project, feeding directly into the design of production high-assurance environments Help scope and prioritize the next wave of Labs projects based on what the current round uncovers You May Be a Good Fit If You Have: 7+ years of software or security engineering experience, with a solid foundation in production systems Some of that time spent on pilots, prototypes, or applied research work where shipping a working answer to a hard question was the explicit goal Strong programming skills in Python and at least one systems language (Go, Rust, or C/C++) Hands-on experience with cloud infrastructure (AWS, GCP, or Azure), Kubernetes, and networking fundamentals sufficient to stand up and tear down isolated environments quickly A track record of cross-functional execution: you can walk into a room with ML researchers, infrastructure engineers, and vendors and leave with a shared plan Clear written communication: you know how to turn six weeks of experimentation into a two-page memo someone can act on Comfort with ambiguity and iteration, having run experiments that failed, extracted the lesson, and moved forward Genuine curiosity about what it would actually take to defend against a nation-state-level adversary Passion for AI safety and a real understanding of the role security plays in making frontier AI development go well Bachelor's degree in Computer Science, a related field, or equivalent industry experience required. Strong Candidates May Also Have: Prior experience in offensive security, red teaming, or security rese