AI Red Team Tester

ixolabs.ai; France; Sur place; EUR 60 000 - 90 000; Temps partiel

Résumé du poste

ixolabs.ai is seeking an AI Red Team Tester to creatively challenge AI systems, uncover vulnerabilities, and improve safety. The role focuses on developing novel attacks against leading models and documenting failure modes in a rigorous, ethical framework.

Ideal candidates have hands-on experience with GPT-3/4, Claude, or Llama, a strong attacker mindset, and excellent written reporting skills. Flexible, part‑time commitment with a 6‑month horizon and potential extension.

Qualifications

• Demonstrated creative problem-solving and an attacker mindset.
• Extensive experience interacting with and probing large language models (e.g., GPT-3/4, Claude, Llama).
• Familiarity with adversarial machine learning concepts, prompt injection, and data poisoning.
• Strong ethical compass and understanding of responsible disclosure principles.
• Excellent written communication skills for reporting complex findings.

Responsabilités

• Develop and execute novel adversarial attacks and jailbreaks against AI models.
• Identify and document failure modes related to harmful content, privacy leakage, and model manipulation.
• Craft sophisticated prompts and multi-turn dialogues to elicit biased or unsafe responses.
• Provide detailed, structured feedback on vulnerabilities, including exploits and mitigations.
• Collaborate with safety researchers to refine testing methodologies and safety benchmarks.
• Stay informed on emerging AI attack vectors and ethical hacking principles.

Connaissances

Attacker mindset

LLM probing

Adversarial ML

Ethical disclosure

Communication

Description du poste

To build truly resilient AI, we must first understand its weaknesses. As an AI Red Team Tester, you’ll embody the adversarial mindset, creatively challenging AI systems to uncover vulnerabilities, push their boundaries, and ultimately make them safer and more robust against malicious use.

Key Responsibilities

• Develop and execute novel adversarial attacks and 'jailbreaks' against cutting‑edge AI models, particularly LLMs.
• Identify and document failure modes related to harmful content generation, privacy leakage, and model manipulation.
• Craft sophisticated prompts and multi‑turn dialogues to elicit biased, toxic, or otherwise undesirable AI responses.
• Provide detailed, structured feedback on AI system vulnerabilities, including potential exploits and mitigation strategies.
• Collaborate with AI safety researchers to refine testing methodologies and contribute to the development of new safety benchmarks.
• Stay informed on emerging AI attack vectors, adversarial machine learning techniques, and ethical hacking principles.

Ideal Qualifications

• Demonstrated creative problem‑solving and an 'attacker's mindset'.
• Extensive experience interacting with and probing large language models (e.g., GPT-3/4, Claude, Llama).
• Familiarity with concepts in adversarial machine learning, prompt injection, and data poisoning.
• Strong ethical compass and understanding of responsible disclosure principles.
• Excellent written communication skills for reporting complex findings.
• Background in cybersecurity, offensive security, or a related field is highly desirable.

Project Timeline

• Start Date: Within 1 week
• Duration: 6 months (with potential for extension)

Commitment

• Flexible, 20‑40 hours per week

Challenge the limits of AI – become an AI Red Team Tester!