AI Red Teamer (Remote)

Written by the T&S Network from Handshake's posting.

Adversarial testing of large language models across many risk categories, including content safety, CBRN, cybersecurity, persuasion, child safety, self-harm, and regulatory compliance. The work finds where model guardrails break so issues are caught before users see them. It is a generalist role spanning text, image, voice, and agentic model capabilities depending on project needs.

• Child safety and grooming
• Self harm and suicide
• Hate speech
• Adult and sexual content
• AI misuse and jailbreaks

What you'd do

• Write creative prompts and multi-turn conversations that probe guardrails across many risk areas
• Run jailbreak, evasion, and prompt injection techniques to get around safety filters and defenses
• Hunt for edge cases that push models toward banned, harmful, or inaccurate outputs
• Grade model replies using organized harm categories and severity scales
• Document experiments, including what was tried, why, and what it revealed
• Review and refine adversarial prompts written by other team members
• Help build out harm categories, run calibration sessions, and check consistency between raters

What it asks for

• Strong hands-on experience using multiple LLMs such as ChatGPT, Claude, Gemini, and open-source models
• Intuition for adversarial prompt crafting
• Clear written communication for experiment documentation
• Familiarity with jailbreak or evasion techniques (preferred)
• Python or other scripting language familiarity (preferred)
• Experience with LLM APIs or evaluation tooling (preferred)

Worth knowing

• Regular and deliberate exposure to violent, self-harm, hate, sexually explicit, and child safety content
• Support resources are available for content exposure
• Hourly W-2 role employed through TCWGlobal

Terms in this role

Red teaming.

Testing a system, often an AI model, by deliberately trying to make it fail or misbehave, so problems are found and fixed first.

Jailbreak.

A prompt written to get an AI model to ignore its safety rules.

Our best summary of the posting, and it can be wrong. The employer's posting is what counts: read it in full on Handshakenorth_east before you apply. If we got something wrong, report this listing.