How to apply for Research Engineer - Neo

About SASH

SASH is a Singapore-based nonprofit that builds AI safety research capacity and governance expertise across Asia. Neo Research (新衡) is its independent AI safety research organization studying frontier risks in capable AI systems, with focus on open-weight models and Asia's frontier-model ecosystem. The work produces empirical evidence about risks that are important but difficult to measure.

About the role

Research engineers build the experimental systems needed to evaluate increasingly capable models: realistic agent environments, model and tool integrations, long-horizon evaluation infrastructure, and systems for reliable, reproducible experiments. The role supports research on misalignment, loss of control, and harmful manipulation, including whether models pursue unintended objectives, conceal behaviour, adapt to evaluations, or evade oversight under greater autonomy.

A typical day

A typical day might involve building or debugging an agent environment, integrating tools for a model evaluation, or running long-horizon experiments and checking their reproducibility. You would coordinate remotely with researchers in Singapore to refine experimental systems for studying misalignment, loss of control, or harmful manipulation. The work is hands-on engineering in service of empirical safety research.

Who SASH is looking for

• Experience building agent environments or evaluation infrastructure for language models, including tool integrations and long-horizon experiments
• Comfort with running experiments that require reliability and reproducibility, and debugging complex model behaviours
• Interest in frontier AI safety topics such as misalignment, loss of control, harmful manipulation, evaluation awareness, and oversight evasion
• Ability to work in a remote full-time role while coordinating with a Singapore-based research organization

Tips for this application

• Read Neo Research's public work on open-weight models and Asia's frontier-model ecosystem, then reference specific questions or methods in your application.
• Describe concrete systems you have built for model evaluation, tool use, or agent environments, including how you handled reliability and reproducibility.
• Explain why open-weight models and the Asian frontier-model ecosystem matter for frontier safety, since the team focuses there.
• Note that the role is remote but the organization is based in Singapore; clarify your working hours and any experience collaborating across time zones.
• Avoid generic AI safety statements. Instead, connect your engineering work to specific risks like evaluation awareness, oversight evasion, or unintended objectives.

What to cover in your cover letter

How you have built experimental systems for evaluating capable models, including agent environments or tool integrations. Your approach to making complex experiments reliable and reproducible. Your understanding of frontier safety risks in open-weight and Asian frontier models. Why you want to work at Neo Research specifically, given its focus on empirical evidence about hard-to-measure risks.

Research before applying

• Read Neo Research's published work or public descriptions of their current focus on misalignment, loss of control, and harmful manipulation.
• Learn which open-weight models are developed in Asia and how they are deployed globally, since the team focuses on this ecosystem.
• Understand SASH's role as a nonprofit building AI safety capacity and governance expertise across Asia.
• Look for any public talks, papers, or blog posts from Neo Research staff to understand their empirical approach and technical stack.

Likely interview topics

Based on the job description, expect questions about:

• How you would design a realistic agent environment to test whether a model conceals problematic behaviour or adapts to evaluations.
• Your experience with model and tool integrations, and how you ensure long-horizon experiments remain reproducible.
• How you would measure misalignment or loss of control in an open-weight model from Asia's frontier ecosystem.
• A time you debugged a complex experiment where model behaviour was unreliable or hard to interpret.
• How you would handle infrastructure for running experiments that may require significant compute or coordination across remote team members.

Common mistakes to avoid

• Submitting a generic AI safety cover letter that does not mention open-weight models, Asia's frontier ecosystem, or the specific risks Neo Research studies.
• Focusing only on model training or benchmarks without addressing agent environments, tool integrations, or long-horizon evaluation infrastructure.
• Ignoring the remote nature of the role or failing to show how you collaborate effectively with a Singapore-based team.