AI Security Researcher/Research Engineer, Brave

• Demonstrated experience working on Secure AI (prompt injection attacks and security guardrails)
• Strong publication record in AI security, ML security, or closely related areas
• A track record of shipping products and features inside a fast-moving environment
• Strong coding skills in Python and familiarity with ML frameworks like PyTorch or JAX
• Experience training, fine-tuning, and evaluating large language models
• Deep familiarity with large language models and web agents, how they work, and how they are trained
• Experience applying security principles (e.g., information flow control, access control/least privilege)
• Demonstrated success in bringing clarity and ownership to ambiguous technical problems
• A Ph.D. in computer science is highly preferred, but we may consider some exceptions with industry experience

Desirable

• We are looking for a researcher/research engineer to help mitigate the security risks that come with building AI systems-especially LLM-powered web agents and chatbots
• In this role, you'll design and implement secure-by-design architectures, audit our LLM-powered web agent (https://brave.com/blog/ai-browsing/) and chatbot, and help set Brave's direction on how we identify, measure, and reduce security risk across AI products
• Design and execute adversarial evaluations to uncover failure modes in LLM-powered web agents and chatbots (e.g., prompt injection, tool abuse, data exfiltration)
• Analyze model behavior (including internal thoughts) to understand and forecast security risks in frontier models
• Develop and prototype security-focused training and fine-tuning techniques
• Augment web agents with security principles including information flow control, privilege control and contextual security
• Partner with engineers and (privacy, ML, and Cryptographic) researchers to translate findings into product changes, tests, and measurable security improvements
• Help define a forward-looking security strategy and policy for building and responsibly releasing web agents—advocating for user protection