Technical Product Manager, Trust & Safety

This is a fully remote position, open to applicants in United States.

📋 Description

• Define and deliver data products for frontier labs focused on training and assessing AI models for personalization, computer usage, and GDP validation tasks.

• Convert lab research objectives into product specifications for safe training data, reinforcement learning environments, and evaluations.

• Create sandboxed reinforcement learning environments featuring misalignment traps, cyber traps, honeypots, and other adversarial scenarios.

• Establish safe-behavior criteria, verifiers, and reward signals for agentic tasks.

• Determine quantifiable safety and trust metrics, including attack success rates, over-refusal incidents, trap trigger rates, and task completion under adversarial conditions.

• Identify blind spots and promote product iteration.

• Frame challenges, evaluate business and technical tradeoffs, and differentiate between minimum viable product and ideal state.

• Collaborate with lab researchers, machine learning and platform engineers, red-team specialists, policy experts, and project leads overseeing expert annotator pools.

• Monitor agentic risks, alignment research, and emerging attack methods and translate them into roadmap initiatives.

• Draft product specifications, customer briefs, and safety findings for researchers and executives.

• Develop cyber-trap environment suites, integrity-trap tasks, misalignment evaluation harnesses, and trap libraries with quality assurance frameworks.

⛳️ Requirements

• Over 5 years of product management experience, particularly with technical products in machine learning, AI data, security, or trust & safety.

• Practical knowledge of LLM training and evaluation methodologies, including supervised fine-tuning, reinforcement learning from human feedback, reward models, verifiers, and benchmark design.

• Proficient understanding of agentic AI, tool utilization, computer-use agents, multi-step planning, and associated failure modes.

• Solid comprehension of AI safety principles such as reward hacking, deceptive alignment, jailbreaks, and prompt injection.

• Technical capability to read code, evaluate environment specifications, and make informed technical decisions.

• Experience in designing metrics that measure risk, system performance, and user impact.

• Proven track record of launching 0-to-1 products in dynamic and uncertain environments.

• Ability to articulate complex technical concepts clearly to researchers, clients, and executives.

• Innovative, adversarial mindset with the capacity to foresee model shortcuts and risks.

• Familiarity with Python and sandboxed environments (Docker, VMs, browser automation) is an advantage.

🏝️ Benefits

• Opportunity to work at the forefront of AI technology.

• Collaborate with a diverse team of experts in various fields.

• Professional development through challenging and impactful projects.

• Flexible working environment.