Research Scientist (AI Safety)

TL;DR

Research Scientist (AI Safety): Studying how LLM agents fail in the wild and eliciting unsafe behaviors through concrete experiments with an accent on agent misalignment and deception. Focus on developing automated audit agents, pressure-testing frontier models, and building empirical benchmarks for AI safety.

Location: Hybrid in Paris or London. Relocation package available for Paris only.

Salary: $150K – $250K + Equity

Company

AI Safety company building a reliability and optimization layer for AI systems using natural-language policies.

What you will do

• Own end-to-end research projects from hypothesis formulation to falsifiable experiments and results.
• Develop automated audit agents to discover and characterize suspect model behavior at scale.
• Study misalignment and bias in real user interactions to create shippable evals.
• Pressure-test frontier agents in high-stakes scenarios to identify failure modes.
• Perform white-box and black-box investigations of AI model failures.
• Publish findings in public blog posts and conference papers.

Requirements

• Proven track record of empirical research in agent behavior, model evaluation, or alignment.
• Strong ML engineering skills, including independent ability to build MVPs with fine-tuning and agent inference.
• Expertise in experimental design, including isolating failure modes and calibrating baselines.
• Ability to define and iterate experiments for vague behavioral questions.
• Fluent AI power-user experienced with frontier models and coding agents.

Nice to have

• Publications at A* venues (NeurIPS, ICML, ICLR, ACL).
• Depth in interpretability (NLAs, SAEs, persona vectors).
• MSc or PhD in ML, Computer Science, Physics, or related quantitative fields.
• AI safety fellowship (MATS, ASTRA, etc.).

Culture & Benefits

• Paid time off according to local regulations.
• Comprehensive medical insurance for France-based employees.
• Provision of all necessary hardware, tools, and AI agent/IDE subscriptions.
• Bi-annual team off-sites (e.g., Alps, Saint-Tropez).

Hiring process

• Introductory HR call (25 min).
• Take-home technical test.
• Technical interview with the Head of Fundamental Research (60 min).
• Final conversation with the CEO (45 min).