Data Science - Agentic AI Intern

Cobalt was founded on the belief of a fundamental human aspiration: the desire to live better and safer. It all started in 2013, when our founders realized that pentesting can be better. Today our diverse, fully remote team is committed to helping organizations of all sizes with seamless, effective and collaborative Offensive Security Testing that empower organizations to OPERATE FEARLESSLY and INNOVATE SECURELY.

Our customers can start a pentest in as little as 24 hours and integrate with advanced development cycles thanks to the powerful combination of our SaaS platform coupled with an exclusive community of testers known as the Cobalt Core. Accepting just 5% of applicants, the Cobalt Core boasts over 400 closely vetted and highly skilled testers who jointly conduct thousands of tests each year and are at the forefront of identifying and helping remediate risk across a dynamically changing attack surface.

Cobalt is an Equal Opportunity Employer and we strive to build a diverse and inclusive workforce at our company. At Cobalt we aspire to engage with diverse individuals, communities, and organizations in order to continue to nurture our unique rich diverse culture. Join our team, and be your true self to do your best work. 

We are pushing the boundaries of agentic AI in the offensive security space, leveraging over 10 years of real-world offensive security data to train and evaluate our models. Our work sits at the intersection of applied mathematics, statistical modeling, and autonomous AI systems, focused on high-stakes, adversarial environments.

We are seeking a graduate-level Data Science Intern with a strong foundation in applied mathematics or statistics and hands-on experience with agentic AI systems. In this role, you will contribute to the development of autonomous agents, improving their ability to reason, adapt, and exploit findings in both test and real-world environments. You will also develop models that capture complex system interactions and evaluate how agents perform under evolving platform conditions.

• Build and improve autonomous agents, focusing on reasoning, planning, and exploitation workflows
• Enhance agent performance in both simulated and real-world environments
• Analyze outputs from agents to identify system weaknesses and emergent behaviors
• Collaborate with engineering and security teams to integrate models and agents into real-world offensive security workflows
• Apply techniques from:
• causal inference
• probabilistic modeling
• optimization

• Design and run experiments to validate model and agent assumptions
• Work with large-scale datasets to calibrate and test models
• Iterate on agent behavior based on empirical results and failure analysis
• Communicate findings clearly to technical stakeholders

• Graduate student (MS/PhD) in a quantitative field
• Strong foundation in at least one of: applied math, probability, statistics, and modeling
• Proficiency in Python
• Experience with agentic LLM-based AI systems

• Familiarity with offensive security concepts
• Experience with causal inference, simulation, or RL/planning
• Research, publications, or open-source contributions

• Ownership of a high-impact, technically challenging project
• Exposure to real-world applications of agentic AI 
• Mentorship from experienced researchers and engineers
• Opportunity to contribute to systems at the intersection of data science, AI, and security