About the Project
An international project in AI Security focused on developing environments that help LLM models undergo reinforcement learning for cybersecurity-related tasks.
These environments will include tasks in various cybersecurity domains (pentesting, incident response, exploitation, malware, etc.). The goal is to create tasks that are complex enough to cause model errors but still solvable by professional specialists.
Important: This is not a classic Pentester or Red Team Engineer position. The specialist's main task is to create complex, but realistic and verifiable security scenarios where LLM models make mistakes specifically due to reasoning deficiencies, rather than due to ambiguous instructions or environment limitations.
Responsibilities
- Design security scenarios for evaluating and training LLM models
- Create environments using GitHub, Slack, and other tools found in real development
- Develop tasks requiring complex multi-step reasoning
- Simulate real offensive security scenarios (e.g., supply chain attacks, malicious pull requests, exploitation via GitHub Actions, and other modern techniques)
- Test developed scenarios on several frontier models and analyze the causes of their errors
- Formulate expected model behavior and describe why a specific error is considered a reasoning failure
- Automate task creation and verification using Python and other tools
What We Expect
- 5+ years of practical experience in Offensive Security, Application Security, or Web Penetration Testing
- Practical experience in AI Red Teaming / LLM Security
- Deep understanding of modern attack techniques and the attack surface of web applications, APIs, GitHub, CI/CD, and related ecosystems
- Practical experience in penetration testing and application security research
- Experience developing in Python and/or JavaScript
- Skills in static and dynamic code analysis (Python, JavaScript, Java)
- Ability to independently research new areas, design complex scenarios, and explain the results of your work
- Fluent English (Upper-Intermediate / B2+)
- Practical experience working with modern LLM models
- Understanding of their operational specifics, limitations, and typical failure modes
- Interest in AI Security and LLM behavior research
- Experience with Agentic AI, AI Agents, or Model Evaluation
Will Be a Big Plus
- Experience researching security of GitHub Actions, CI/CD, Supply Chain Security
- OSWE, eWPTXv2, OWEE, OSWA certificates or similar
- Publications, write-ups, participation in Bug Bounty, HackTheBox, TryHackMe, Pwn2Own, or other research projects
The most valuable competence here is not the ability to exploit the next XSS, but the ability to invent a new, unique scenario that breaks a strong model.
Here, research mindset, creativity, and the ability to invent new attack scenarios are important, rather than replicating already known techniques.
Conditions
- B2B contract (with sole proprietorship)
- Payment in US dollars
- Part-time/Full-time: hourly payment for completed tasks, 1 task takes about 10-12 hours
- Location — strictly outside of Russia/Belarus
