Research Engineer

Confero; San Francisco (CA); Hybrid; USD 200,000 - 300,000; Full time

Job summary

Confero in San Francisco is seeking a Research Engineer, AI Safety / Alignment to join a hybrid team working three days in the office. You will design and build evaluation environments and RL-style tests to study how frontier models behave when optimized for the wrong objective.

You will own projects end to end, from scenarios to agents and graders, and contribute to infrastructure that runs and evaluates these experiments.

Qualifications

• Strong generalist software engineering ability with Python preference.
• At least 1 year of professional software engineering experience.
• Genuine interest in AI alignment and reducing existential risk.
• Strong analytical and conceptual reasoning.
• Comfortable owning ambiguous technical projects end to end.

Responsibilities

• Design and build evaluation and RL-style environments to test frontier models for reward hacking.
• Own research projects end to end: developing scenarios, building environments, directing agents, iterating graders, analyzing behavior.
• Contribute to shared infrastructure used to create, run and evaluate these environments.

Skills

Python

Software engineering

Analytical thinking

Project ownership

AI alignment interest

Tools

RL frameworks

Job description

Research Engineer, AI Safety / Alignment

San Francisco | Hybrid, 3 days in office

$200K-$300K base + equity

We’re building a new applied AI research team focused on AI alignment and existential risk, studying how increasingly capable models and agents behave when given opportunities to exploit, cheat or optimise for the wrong objective.

What you’ll do

You’ll design and build evaluation and RL-style environments that test frontier models for behaviours such as reward hacking and exploiting weaknesses in tasks or graders.

You’ll own research projects end to end: developing scenarios, building environments, directing agents, iterating graders, analysing model behaviour and refining the experiment.

You’ll also contribute to the shared infrastructure used to create, run and evaluate these environments.

Prior experience with RL environments, agent evaluations or reinforcement learning is valuable, but not required.

A good fit:

• Strong generalist software engineering ability, ideally Python
• 1+ years of professional software engineering experience
• Genuine interest in AI alignment and reducing existential risk from advanced AI
• Strong analytical and conceptual reasoning
• Comfortable owning ambiguous technical projects end to end
• Interested in working directly with frontier LLMs and AI agents
• Experience with evals, agents, RL environments, security or adversarial systems is a plus