Department: Intelligence
Location: Remote
Employment Type: Full-time
As a GenAI CBRNE Cyber Red Team Expert, you will play a critical role in adversarially testing and strengthening the safety guardrails of GenAI systems against high-risk cyber threats involving CBRNE domains.
You will combine deep cybersecurity expertise with CBRNE domain knowledge and hands-on AI red-teaming to design sophisticated adversarial prompts, multi-turn attack scenarios, jailbreaks, and model evaluations. Your work will identify circumstances in which GenAI systems could inadvertently provide information or capabilities that materially enable cyber-enabled CBRNE threats.
The role requires an adversarial mindset: thinking creatively about how malicious actors could manipulate AI systems, combine seemingly benign information across multiple interactions, circumvent safeguards, or exploit model behavior to obtain sensitive cyber-CBRNE information.
Execute rigorous adversarial red-teaming of Generative AI models across cyber-CBRNE threat scenarios, systematically probing models for vulnerabilities, safeguard bypasses, dangerous capability escalation, and unintended disclosure of sensitive operational information.
Design and execute adversarial prompts, jailbreaks, prompt mutations, multi-turn conversations, and scenario-based evaluations that test whether model safeguards remain effective against sophisticated or obfuscated cyber-CBRNE requests.
Develop realistic cyber-CBRNE attack scenarios involving critical infrastructure, industrial control systems, operational technology, cyber-physical systems, laboratory environments, hazardous-material facilities, and other high-consequence systems.
Evaluate whether models can be manipulated into materially assisting threat actors through attack planning, vulnerability analysis, target-specific reasoning, operational troubleshooting, or the aggregation of individually benign information into higher-risk workflows.
Identify and document failure modes and attack patterns, including indirect requests, role-playing, encoded or obfuscated prompts, terminology substitution, decomposition of harmful objectives into benign-looking subtasks, and multi-turn escalation.
Conduct systematic taxonomy audits and safety evaluations, classify model failures by severity and exploitability, reproduce findings, and provide actionable recommendations to AI safety and model-alignment teams.
Develop repeatable red-team test suites, adversarial datasets, evaluation rubrics, and risk taxonomies for measuring model resilience against emerging cyber-CBRNE threats.
Maintain current knowledge of emerging GenAI attack techniques, AI safety research, cyber threat intelligence, CBRNE security risks, and cybersecurity threats affecting critical infrastructure and high-consequence environments.
Must-Have:
Location: Must be located in and authorized to work within the United States (excluding Illinois and Texas) or United Kingdom.
Education: Advanced degree in Cybersecurity, Computer Science, Engineering, Security Studies, CBRNE-related sciences, or a closely related technical field. Equivalent advanced professional, military, intelligence, government, or industry experience may be considered.
Cybersecurity Expertise: Deep understanding of cybersecurity concepts, adversarial techniques, vulnerability analysis, attack chains, threat modeling, and defensive security.
CBRNE Knowledge: Strong understanding of security and risk considerations associated with Chemical, Biological, Radiological, Nuclear, and/or Explosive environments, particularly their intersection with cyber and digital systems.
Red Team Expertise: Demonstrated experience with red teaming, penetration testing, adversarial simulation, vulnerability research, security testing, threat emulation, or comparable offensive-security methodologies.
GenAI Red Teaming: Experience or demonstrated aptitude in adversarial prompt generation, jailbreak research, prompt mutation, multi-turn testing, model behavior analysis, and evaluation of LLM safety controls.
ICS/OT Knowledge: Familiarity with industrial control systems, SCADA, operational technology, cyber-physical systems, or critical-infrastructure environments.
Adversarial Mindset: Ability to think creatively about how sophisticated users could circumvent model safeguards through decomposition, obfuscation, contextual manipulation, multi-turn interactions, or combinations of otherwise permissible information.
Communication: Strong technical writing skills with the ability to clearly document prompts, attack methodology, model responses, vulnerabilities, reproduction steps, severity assessments, and recommended mitigations.
Alice.io, formerly known as ActiveFence, is a trust, safety, and security company specializing in safeguarding artificial intelligence systems across their entire lifecycle. The company focuses on identifying and mitigating risks in generative AI through domain expertise in Chemical, Biological, Radiological, and Nuclear (CBRN) areas. Alice.io offers coverage and security solutions designed to address the emerging challenges posed by AI technologies. The company operates an 'Elite Collective' model, engaging specialized experts on a project-based, on-demand basis to conduct high-impact interventions for sophisticated technology challenges.
