How to apply for Lead Researcher, AI Offense-Defense Dynamics

About Center for AI Risk Management and Alignment

The Center for AI Risk Management and Alignment (CARMA) is a nonprofit dedicated to improving the technical, policy, and multidisciplinary aspects of AI risk management, safety, and alignment. Unlike many AI safety organizations, CARMA explicitly bridges technical research with policy intervention design, making it a hub for those who want to translate risk analysis into actionable governance. Working here means contributing to a mission-driven team focused on some of the most pressing challenges in AI, with a global remote-first culture.

About the role

As the Lead Researcher for AI Offense-Defense Dynamics, you will spearhead a novel research agenda examining how AI capabilities shift the balance between offensive and defensive applications, with the goal of informing safer AI development and policy. You'll develop quantitative system dynamics models, expand a taxonomy of offense-defense dynamics, and design analytical tools to identify policy levers that steer AI toward defensive, safety-enhancing uses. This role is pivotal in shaping how governments and organizations anticipate and mitigate AI-enabled risks.

A typical day

A typical day might involve refining a system dynamics model of AI offense-defense interactions, collaborating with policy analysts to identify intervention points, and drafting a section of a taxonomy paper. You'll also attend virtual team meetings to discuss research priorities and may present findings to stakeholders or policymakers. Expect a mix of independent deep work and interactive problem-solving with a distributed, interdisciplinary team.

Who Center for AI Risk Management and Alignment is looking for

• Has a strong background in AI safety, risk management, or a related field, with demonstrated experience in quantitative modeling and system dynamics—ideally applied to technological risk or security domains.
• Possesses a track record of developing taxonomies and metrics for complex systems, such as classifying AI capabilities or risk factors, and can translate these into predictive models.
• Is skilled in policy analysis and intervention design, able to connect technical findings to actionable governance strategies and communicate effectively with policymakers.
• Thrives in a remote, interdisciplinary environment and is eager to lead research that bridges technical, social, and institutional dimensions of AI risk.

Tips for this application

• Highlight any experience you have with offense-defense dynamics, even if from adjacent fields like cybersecurity, military strategy, or arms control—CARMA values cross-domain expertise.
• Showcase quantitative modeling skills by including a portfolio or examples of system dynamics models you've built, especially those involving technological or social factors.
• Demonstrate familiarity with CARMA's existing work: reference their publications or taxonomy in your application to show alignment with their research direction.
• Emphasize your ability to design policy interventions: describe specific instances where your analysis informed a policy recommendation or decision.
• Since the role is remote and global, underscore your experience working in distributed teams and your ability to lead independent research across time zones.

What to cover in your cover letter

Your cover letter should emphasize: (1) your specific interest in offense-defense dynamics and why you see it as critical for AI safety; (2) your quantitative modeling expertise and how you would apply it to build system dynamics models for AI risk; (3) any experience developing taxonomies or metrics for complex systems; and (4) your capacity to translate research into policy interventions, ideally with examples. Also mention why CARMA's nonprofit, multidisciplinary approach appeals to you.

Research before applying

• Read CARMA's published reports and blog posts, especially any that touch on offense-defense dynamics or AI risk taxonomies, to understand their current framing and gaps.
• Explore the backgrounds of CARMA's team and advisors to identify potential collaborators and to tailor your application to their expertise.
• Investigate existing literature on offense-defense balance in other domains (e.g., cybersecurity, international relations) to draw parallels and propose novel adaptations.
• Familiarize yourself with key AI policy debates and governance frameworks (e.g., EU AI Act, US executive orders) to ground your policy intervention ideas in current contexts.

Likely interview topics

Based on the job description, expect questions about:

• How would you define and measure 'offense-defense dynamics' in the context of advanced AI? Walk us through a taxonomy you might develop.
• Describe a system dynamics model you've built. What variables did you include, and how did you validate it?
• Can you give an example of a policy intervention that could shift the offense-defense balance toward safer outcomes? How would you model its impact?
• What are the biggest challenges in predicting whether an AI capability will be used offensively or defensively? How might you address them?
• How would you collaborate with CARMA's policy and technical teams to ensure your research is actionable?

Common mistakes to avoid

• Being too generic: avoid vague statements about 'wanting to make AI safe' without tying them to offense-defense dynamics or CARMA's specific research areas.
• Neglecting the policy component: since the role requires policy analysis and intervention design, focusing only on technical modeling without addressing governance implications is a mistake.
• Ignoring the nonprofit context: candidates who emphasize profit-driven or purely academic motivations may not align with CARMA's mission-driven, multidisciplinary approach.