How to apply for AI Red Teamer

About Trajectory Labs, PBC

Trajectory Labs is a startup that builds reinforcement learning environments for frontier AI labs. Its work focuses on training models that are robust, secure, and reliable. This role sits at the point where AI safety research meets practical attack testing.

About the role

As an AI red teamer, you will probe frontier AI systems for vulnerabilities and design novel attack strategies. You will run jailbreak attempts against state-of-the-art models, identify security failures, and help build evaluation infrastructure and safety benchmarks. The work directly informs how frontier labs harden their models.

A typical day

A typical day might involve running attack scripts against a frontier model, logging which prompts succeed or fail, and updating an evaluation harness. You would also spend time designing new attack strategies and reviewing results with the team. The exact rhythm depends on the current benchmark or model target, so ask about team workflow and release cycles in the interview.

Who Trajectory Labs, PBC is looking for

• Has demonstrated success jailbreaking frontier models, with concrete examples or writeups to show.
• Is self-directed and comfortable defining their own attack strategies without waiting for detailed instructions.
• Is proficient in Python scripting and command-line workflows.
• Has hands-on experience with agentic coding tools and can use them to scale red-teaming work.

Tips for this application

• Lead with specific jailbreak results. Name the models, the failure modes, and what you got the model to do.
• Include links to public writeups, repos, or benchmark contributions that show your attack methodology.
• Show how you use agentic coding tools in your workflow, since the role values that expertise.
• Explain your evaluation infrastructure experience, even if it is small-scale, because the role includes contributing to benchmarks.
• Keep the application focused on attack and evaluation work rather than general AI safety interest.

What to cover in your cover letter

Describe one or two jailbreaks you pulled off and why they mattered. Explain how you design novel attack strategies rather than reusing known prompts. Show that you can build or extend evaluation infrastructure, not just run tests. State why you want to work on RL environments for frontier labs specifically.

Research before applying

• Read Trajectory Labs' website and any public material on the RL environments they build.
• Look into how frontier AI labs use red-teaming and evaluation infrastructure in their safety work.
• Study recent public jailbreak techniques and benchmark papers to understand the current attack landscape.
• Check whether Trajectory Labs has published anything about their methods, clients, or past projects.

Likely interview topics

Based on the job description, expect questions about:

• Walk through a specific jailbreak you developed, including your process and what failed.
• How would you design an evaluation benchmark to measure a model's resistance to a new attack class?
• What agentic coding tools do you use, and how do they change your red-teaming workflow?
• How do you decide when a vulnerability is worth reporting versus when it is noise?
• How would you approach red-teaming a model you have never seen before, with no prior attack surface knowledge?

Common mistakes to avoid

• Claiming red-teaming experience without showing actual jailbreak examples or artifacts.
• Treating the role as generic AI safety work instead of hands-on adversarial testing.
• Focusing only on prompt tricks while ignoring evaluation infrastructure and benchmark contributions.