AI Red Teamer (LLM Generalist)

About the role

Stress-test large language models by trying to break them. Your findings support the safety of models from leading AI labs. Full-time hours (40 h/week), W-2 hourly through TCWGlobal; you must live in or relocate to the Seattle area.

What you'll do

• Design adversarial prompts and multi-turn scenarios across risk categories (content safety, cybersecurity, persuasion and more).
• Probe guardrails with jailbreak, evasion and prompt-injection techniques.
• Score responses against harm taxonomies and document every experiment clearly.

Good to know

The work involves regular exposure to disturbing content.