Adversarial Prompt Testing Specialist

Globik AI; United States; Remote; USD 17,000 - 25,000; Full time

Job summary

Globik AI is seeking Adversarial Prompt Testing Specialists to stress-test language models with edge-case prompts and strategies. You will craft prompt variations, log model outputs, and categorize failures to aid model improvement.

Responsibilities include testing paraphrases, multi-turn prompts, and tracking regressions across model versions. You will provide structured feedback with reproducible examples and maintain a log of effective adversarial patterns.

Qualifications

• Experience crafting adversarial and boundary-pushing prompts.
• Ability to test robustness across paraphrases, obfuscation, and multi-turn setups.
• Ability to log model outputs and flag inconsistent or unstable behavior.

Responsibilities

• Construct adversarial and boundary-pushing prompt variations for target models.
• Test prompt robustness across paraphrases, obfuscation, and multi-turn setups.
• Log model outputs and flag inconsistent or unstable behavior.
• Categorize failure modes (instruction leakage, contradiction, over-compliance, etc.).
• Compare responses across model versions to track regressions.
• Provide structured feedback with reproducible examples.
• Maintain a running log of effective adversarial patterns.

Skills

Adversarial prompts

Prompt engineering

Tools

Logging tools

Job description

Globik AI is seeking Adversarial Prompt Testing Specialists to stress-test language models with edge-case, ambiguous, and manipulative prompts to surface weaknesses in model robustness. You will construct prompt variations, log model behavior, and categorize failure modes for model improvement.

Responsibilities

• ey Responsibilities


• Construct adversarial and boundary-pushing prompt variations for target models.


• Test prompt robustness across paraphrases, obfuscation, and multi-turn setups.


• Log model outputs and flag inconsistent or unstable behavior.


• Categorize failure modes (instruction leakage, contradiction, over-compliance, etc.).


• Compare responses across model versions to track regressions.


• Provide structured feedback with reproducible examples.


• Maintain a running log of effective adversarial patterns.

Pay

$12-$18/task