Job description
Trajectory Labs, PBC • Remote
About us
Our mission is to automate AI safety, to pave the way for a future where the vast majority of AI safety work is done by AI models.
Frontier models already solve coding problems that take humans days, but a model that can be hijacked by a malicious email or web page can't be trusted to work on its own. Before AI can do the work that matters, including AI safety research itself, models have to be robust to attack. So we build the safety and alignment evals, red-teaming programs, and RL environments that find these failures and train them out.
Although Trajectory is less than a year old, we have already been featured in Anthropic's Fable 5 & Mythos 5 System Card (the only external team to bypass the cyber safeguards), led Anthropic's prompt injection evaluation of Auto mode, and been featured in Meta's Muse Spark Safety & Preparedness Report.
About the role
You will attack the cyber safeguards of frontier AI models: the guardrails meant to stop them from writing exploits, escalating privileges, and assisting with real intrusions. You'll design novel attack strategies, document what breaks, and help the labs train it out.
The work looks like offensive security applied to a new attack surface. Your findings shape system cards and deployment decisions for the most capable AI systems.
About you
Essential
- Successful jailbreaks of frontier models (GPT-5, Claude, Gemini, etc.)
- Security mindset
- Self-directed: you work independently with minimal guidance
- Creative problem-solving and lateral thinking
Highly valued
- Offensive security experience: pentesting, red teaming, exploit development, vulnerability research, or serious CTF play
- CTF competition results, published CVEs, or a bug bounty track record
- Python scripting and command-line proficiency
- Expertise with agentic coding tools
Application process
- Submit your application with a resume
- Complete Terminal, The Game
- Interview with the founders
- Offer
Applications reviewed on a rolling basis.
Apply now $ /$