AI Red Teamer, CBRNE

Accountabilities

• Design technically grounded adversarial evaluation scenarios that test whether AI models appropriately handle CBRNE-related requests and resist attempts to obtain dangerous information.
• Evaluate model outputs for technical accuracy, real-world significance, and potential harmful uplift, distinguishing genuinely consequential information from superficial or publicly available knowledge.
• Explore dual-use boundaries by testing scenarios that combine legitimate scientific, medical, industrial, or research contexts with potential weapons applications.
• Conduct structured multi-step and multi-turn evaluations to assess whether models can be incrementally manipulated into providing unsafe information.
• Score and categorize model responses using established harm taxonomies, severity frameworks, and real-world risk criteria.
• Document evaluation findings with clear technical reasoning, explaining model strengths, weaknesses, and the significance of identified failures.
• Assess the distinction between open-source information and operationally meaningful knowledge that could materially increase harmful capabilities.
• Contribute to the creation and refinement of CBRNE-specific evaluation frameworks, threat models, taxonomies, and testing methodologies.
• Collaborate with AI researchers, fellow red teamers, policy specialists, and other stakeholders to translate technical findings into actionable safety improvements.
• Monitor developments in AI capabilities, adversarial and jailbreak techniques, and relevant developments across the candidate’s CBRNE specialty.
• Perform all evaluation activities within established ethical, safety, and operational security protocols when handling sensitive CBRNE subject matter.

Requirements

• Graduate-level education or equivalent professional experience in a relevant CBRNE field, such as chemistry, biochemistry, microbiology, virology, nuclear physics, radiochemistry, materials science, chemical engineering, or munitions and ordnance.
• Deep subject-matter expertise enabling accurate assessment of technical content and its potential real-world consequences within at least one CBRNE domain.
• Strong understanding of dual-use research considerations and the distinction between publicly available scientific knowledge and operationally significant harmful uplift.
• Hands-on experience working with multiple large language models, such as ChatGPT, Claude, Gemini, or open-source models.
• Strong creative and adversarial problem-solving abilities, with the capacity to think systematically about attack paths, threat models, and model weaknesses.
• Excellent written communication skills, including the ability to explain complex technical risks clearly to non-specialist audiences.
• Strong ethical judgment, discretion, and the ability to approach adversarial scenarios objectively while maintaining clear professional boundaries.
• Ability to work independently, manage structured evaluation tasks, incorporate feedback, and collaborate effectively with multidisciplinary teams.
• Active or previous security clearance, including Secret, Top Secret, or SCI, is a plus.
• Experience in threat assessment, WMD analysis, intelligence analysis, arms control verification, biosafety/biosecurity, chemical safety, nuclear nonproliferation, or explosive ordnance disposal is preferred.
• Familiarity with regulatory frameworks such as the Chemical Weapons Convention (CWC), Biological Weapons Convention (BWC), IAEA safeguards, ATF regulations, or Export Administration Regulations is advantageous.
• Experience with red teaming, penetration testing, structured adversarial evaluation, trust and safety, content moderation, or AI evaluation is a plus.
• Familiarity with Python, scripting languages, LLM APIs, or evaluation tooling is beneficial.
• Published research or professional presentations in a relevant CBRNE field are valued.

Benefits

• Compensation: $65–$158 per hour, with final compensation determined by factors including location, experience, expertise, and other relevant considerations.
• Work arrangement: Fully remote within the United States.
• Employment type: Contract position.
• Mission-driven work: Opportunity to contribute to AI safety and help strengthen safeguards around highly consequential technologies.
• Collaborative environment: Work alongside AI researchers, security specialists, policy teams, and other technical experts.
• Professional impact: Apply specialized CBRNE expertise to emerging challenges at the intersection of artificial intelligence and national security.
• Structured environment: Work with established evaluation frameworks, ethical guidelines, and operational security protocols.
• Specialized focus: Engage in technically challenging adversarial evaluations designed to identify meaningful model vulnerabilities.