This is an opportunity to lead a research direction within a team focused on projects in the new, dynamically developing field of artificial intelligence security. Your tasks will involve finding and researching vulnerabilities in GenAI and AI applications, including AI agents, and exploiting them. Within a team of experienced experts, you will assess the security of the software being developed, propose tools for automating the research and security testing of GenAI models, and also form and develop a research team, participating in international and internal CTFs.
This is not an administrative role; we are looking for a "player-coach" who can write the first prototype, test a threat hypothesis, and solve a CTF task themselves – and only then delegate scaling to the team. If your experience is limited to process management without hands-on development and research, this position will not be suitable.
You will:
- GenAI threat research: creation, support, and development of tools for analyzing and finding threats, including:
- development of automation tools for collecting, storing, and analyzing new publications, incidents, and attack methods on GenAI (crawling, information parsing, development and management of storage systems)
- creation and testing of prototypes (Proof of Concept) for finding GenAI vulnerabilities (scripts, core GenAI libraries, AI agents for code generation)
- DevOps: CI/CD, version control (git), quality testing, infrastructure for operation (Linux)
- Implementation of AI PDLC practices: research and application of methods and tools for creating and developing software using advanced AI models
- Ecosystem immersion: studying and practical application of advanced frameworks and libraries for working with LLMs and AI agents
- Team management: forming and developing a research team (3–5 people), prioritizing research, defending decisions to management and related teams
- CTF and hackathons: forming and captaining a team at international CTFs on AI/ML security, translating discovered techniques into the direction's work
- Publications and regulatory documents: documenting research results, contributing to AI security regulatory documents.
What is important to us:
- Profile and education: completed higher education in IT, computer science, applied mathematics, software engineering, or related fields
- Advanced programming skills: understanding of software design and development principles, experience in writing and reading program code (understandable, maintainable)
- DevOps skills: experience with Git, Linux administration, automation of software development lifecycle stages
- LLMs and AI agents: expert-level application of LLMs for software development, deep understanding of agent architecture and their attack vectors
- English language proficiency at a level sufficient for reading technical documentation, articles, and participating in international CTFs
- Teamwork experience in development projects: Kanban (sprints, task tracking), Jira/Git/Confluence, presentations and the ability to clearly explain ideas to the team/management and coordinate requirements
- Adversarial ML and LLM red teaming: practical experience in finding vulnerabilities in GenAI models (prompt injection, jailbreaking, tool poisoning, data extraction)
- Knowledge of tools for LLM security: Giskard, Garak, LLM Guard, etc.
- CTFs and hackathons: participation in international AI/ML security CTFs (preferably with wins or prizes).
We offer:
- comfortable modern office at 19 Vavilov Street
- office-based work format
- annual salary review, annual bonus
- corporate gym and recreation areas
- over 400 educational programs from SberUniversity for professional and career development
- adaptation program and mentor support at the start
- expanded voluntary medical insurance, preferential family insurance, and corporate pension program
- flexible mortgage loan discount, equal to 1/3 of the Central Bank's key rate
- Prime subscription with the possibility of sharing with three close contacts
- reward for recommending friends to the Sber team.
