LLM Jailbreak Testing Specialist

Globik AI; United States; Remote; USD 17,000 - 28,000; Full time

Job summary

Globik AI is seeking LLM jailbreak testing specialists to probe language models with adversarial prompts and determine whether safety guardrails can be bypassed. You will design and execute jailbreak attempts, document successful bypass techniques, and rate model responses against safety policy.

Responsibilities include recording outcomes, classifying bypass techniques, rating unsafe content, writing clear reproduction steps, tracking new jailbreak patterns, and maintaining confidentiality of

Qualifications

• Strong understanding of LLM behavior prompting and safety mechanisms.
• Comfortable working with sensitive or policy-violating content in a controlled test setting.
• Prior red-teaming security research or AI safety experience is a plus.

Responsibilities

• Design adversarial prompts intended to bypass model safety guardrails.
• Execute structured jailbreak attempts across assigned model versions.
• Record whether an attempt succeeded, partially succeeded, or failed.
• Classify the bypass technique used (roleplay, encoding, multi-turn escalation, etc.).
• Rate the severity and category of any unsafe content produced.
• Write clear reproduction steps for confirmed vulnerabilities.
• Track novel jailbreak patterns as they emerge across sessions.
• Maintain strict confidentiality of all test prompts and findings.

Skills

LLM prompting

AI safety

Red-teaming

Job description

Globik AI is seeking LLM Jailbreak Testing Specialists to probe language models with adversarial prompts and determine whether safety guardrails can be bypassed. You will design and execute jailbreak attempts, document successful bypass techniques, and rate model responses against safety policy.

Responsibilities

• Design adversarial prompts intended to bypass model safety guardrails.
• Execute structured jailbreak attempts across assigned model versions.
• Record whether an attempt succeeded, partially succeeded, or failed.
• Classify the bypass technique used (roleplay, encoding, multi-turn escalation, etc.).
• Rate the severity and category of any unsafe content produced.
• Write clear reproduction steps for confirmed vulnerabilities.
• Track novel jailbreak patterns as they emerge across sessions.
• Maintain strict confidentiality of all test prompts and findings.

Qualifications

• Strong understanding of LLM behavior prompting and safety mechanisms.
• Comfortable working with sensitive or policy-violating content in a controlled test setting.
• Prior red-teaming security research or AI safety experience is a plus.

• Sign a confidentiality and responsible-disclosure agreement before onboarding.
• Follow the assigned safety taxonomy precisely when classifying findings.
• Participate in calibration sessions to align severity ratings across the team.
• Escalate high-severity findings immediately per project protocol.
• Regular quality audits will be conducted on submitted test cases.
• A reliable internet connection and computer are required.

Pay

$12-$20/task