AnthropicBerkeley, CAFour-month fellowship for researchers and engineers to conduct empirical AI security research, producing public outputs like papers. Requires strong technical background, Python fluency, and interest in AI safety; open to remote in US/UK/Canada.
The Anthropic Fellows Program accelerates AI security and safety research by providing funding and mentorship to promising technical talent for four months. Fellows work on empirical projects using external infrastructure (e.g., open-source models, public APIs) to produce public outputs like papers.
Strong candidates may also have:
Candidates must be:
Expected base stipend is 3,850 USD / 2,310 GBP / 4,300 CAD per week (40 hours/week) for 4 months (possible extension).
Develop tools and frameworks to counter AI-enabled cyber threats, design experiments evaluating autonomous AI cyber capabilities, and translate findings into policy-informing demonstrations. Requires deep cybersecurity expertise, Python skills, and experience with LLM agents.
Build scalable systems and data pipelines for security observability, enhancing detection, forensics, and compliance. Requires strong engineering in Python/Golang, Terraform, Azure, and data pipelines with a generalist SRE mindset.
Red Team Engineer on Anthropic's Safeguards team performing adversarial testing and full kill-chain attacks on AI products to uncover vulnerabilities, jailbreaks, and novel abuse vectors before malicious exploitation. Requires strong pen-testing, red teaming, and LLM-specific security experience.
Hardens and secures the OS layer for Anthropic's AI infrastructure, designing kernel-level protections, runtime enforcement, attestation, and isolation for large-scale model training environments. Requires deep Linux kernel expertise, C/systems programming, and OS security experience.
Build systems to detect and prevent account abuse by analyzing signals at scale, integrating vendor data, and creating monitoring tools. Requires 3+ years experience in software engineering focused on fraud/abuse, proficiency in Python/SQL.