Vulnerability Researcher (AI Security)

Location: Remote within Poland or Bulgaria

Company

Specialized evaluation infrastructure is being built for AI safety and robustness testing, simulating adversarial conditions so AI development teams can validate agent behavior before deployment.

What you will do

• Review cybersecurity prompts that an AI system could not confidently classify as benign or malicious.
• Determine the underlying intent and assess whether each prompt supports a legitimate purpose or potential harm.
• Evaluate technical context, specificity, and the real-world uplift a response could provide to an attacker.
• Annotate prompts with intent labels and concise written rationales according to project guidelines.
• Independently review the same prompts as a second researcher and escalate disagreements to a reviewer.
• Flag ambiguous cases, emerging attack techniques, and gaps in the guidelines to support their improvement.

Requirements

• At least 5 years of hands-on experience in vulnerability research and reverse engineering.
• Proven ability to discover vulnerabilities through fuzzing, static analysis, dynamic analysis, and patch diffing.
• Exploit development experience covering memory corruption, heap exploitation, ROP, and mitigation bypasses including ASLR, DEP, and CFI.
• Fluent use of IDA, Ghidra, Binary Ninja, and debuggers including WinDbg, GDB, and LLDB on x86/x64 and ARM binaries.
• Understanding of the CVE lifecycle, n-day and 0-day work, and the weaponization of public PoCs.
• Strong analytical judgment to infer a user's underlying goal from a short prompt and distinguish defensive, educational, and research use from attack preparation.
• Clear written communication in English and the ability to document reasoning concisely.

Nice to have

• OSED, OSEE, or GIAC GXPN certification.
• Online achievements, write-ups, or contributions on HackTheBox, Pwn2Own, TryHackMe, bug bounty programs, or published security research.

Culture & Benefits

• Full-time remote work within Poland or Bulgaria.
• Independent, detail-oriented, organized work with an emphasis on learning and solving complex problems.
• Work connected to AI safety, robustness testing, and adversarial evaluation.