How to apply for Head of Cyber Safety

About Gray Swan

Gray Swan is an AI security company that builds tools to automatically assess risks in AI models. They evaluate AI models for leading frontier labs and build real-time threat detection and adversarial red teaming agents. The team is about 50 people, well-funded, and growing quickly, with work that directly influences how AI agents and systems are deployed at scale.

About the role

This role is to build and lead Gray Swan's cyber safety capability from the ground up. You will be the technical authority on AI-enabled cyber risk across red-teaming, evaluation, benchmarking, defenses, and safety infrastructure. You will define how frontier AI systems are evaluated for offensive cyber capabilities and partner with leading AI labs to reduce real-world security risks.

A typical day

A typical day might involve designing adversarial evaluations for frontier AI models, reviewing results with AI labs, and building scalable evaluation methodologies or safety infrastructure. You could also lead red teaming exercises, benchmark offensive cyber capabilities, and collaborate with machine learning engineers to automate defenses. Since the role is remote and the team is about 50 people, you would coordinate closely with colleagues across time zones.

Who Gray Swan is looking for

• Deep expertise in offensive cybersecurity, vulnerability research, or adversarial AI security.
• Experience evaluating frontier AI models for cyber capabilities or risks.
• Ability to design and lead adversarial evaluations and build scalable evaluation methodologies.
• Driven to reduce catastrophic cyber risks from increasingly capable AI systems.
• Comfortable working at the intersection of offensive security, AI safety, and machine learning.

Tips for this application

• Highlight specific experience evaluating frontier AI models for offensive cyber capabilities or risks.
• Show concrete examples of leading adversarial red teaming or vulnerability research projects.
• Explain how you have built or scaled evaluation methodologies, benchmarks, or safety infrastructure.
• Demonstrate familiarity with Gray Swan's work: evaluating AI models for frontier labs and building adversarial red teaming agents.
• Emphasize your ability to partner with leading AI labs and define industry standards for frontier model security.

What to cover in your cover letter

['Your deep offensive cybersecurity or vulnerability research background and how it applies to AI-enabled cyber risk.', 'Specific experience evaluating frontier AI models for offensive cyber capabilities.', 'How you have designed or led adversarial evaluations and built scalable evaluation methodologies.', 'Your motivation to reduce catastrophic cyber risks from increasingly capable AI systems and help establish industry standards.']

Research before applying

• Gray Swan's product: tools that automatically assess risks of AI models and real-time threat detection.
• Their work with leading frontier labs on AI model evaluation and adversarial red teaming.
• The company's mission to empower safe and secure use of AI and their focus on catastrophic cyber risks.
• Their team size (about 50), funding status (well-funded), and recent growth or announcements.

Likely interview topics

Based on the job description, expect questions about:

• How you would design an adversarial evaluation for a frontier AI model's offensive cyber capabilities.
• Your approach to building a cyber safety capability from the ground up, including team, processes, and infrastructure.
• Examples of past vulnerability research or red teaming and how you translated findings into scalable evaluations.
• How you would partner with leading AI labs to reduce real-world security risks from frontier models.
• Your view on the intersection of offensive security, AI safety, and machine learning in evaluating AI models.

Common mistakes to avoid

• Focusing only on traditional cybersecurity without connecting it to AI model evaluation or adversarial AI security.
• Failing to show specific experience with frontier AI models or offensive cyber capabilities.
• Giving generic answers about leadership without concrete examples of building a capability from the ground up.