AI Safety Expert — English & Swedish

This contract role offers an opportunity to help identify and address safety risks in conversational AI models and intelligent agents. You will conduct structured red-team testing to uncover jailbreaks, prompt injections, misuse scenarios, and potential bias exploitation. Your work will help surface vulnerabilities and systemic risks before they impact users or real-world deployments. You will generate high-quality human evaluation data by annotating failures and classifying different types of vulnerabilities. The role combines rigorous testing methodologies with creative adversarial thinking and strong analytical judgment. Working remotely and asynchronously, you will produce actionable findings that contribute directly to safer, more robust, and more reliable AI systems.

Accountabilities:

• Conduct red-team evaluations of conversational AI models and agents to identify jailbreaks, prompt-injection vulnerabilities, misuse cases, and potential bias exploitation.
• Develop creative adversarial prompts and scenarios to systematically probe model behavior and uncover weaknesses.
• Generate high-quality human evaluation data by annotating model failures, classifying vulnerabilities, and flagging systemic safety risks.
• Apply established taxonomies, benchmarks, frameworks, and testing playbooks to ensure consistent and rigorous evaluations.
• Document findings in a reproducible manner through detailed reports, datasets, attack cases, and supporting evidence.
• Communicate identified risks and technical findings clearly to both technical and non-technical stakeholders.
• Adapt testing strategies across different AI systems, projects, use cases, and requirements while maintaining consistent evaluation standards.
• Work independently and asynchronously while meeting deadlines and maintaining high-quality evaluation standards.

Requirements:

• Fluent or professional-level proficiency in both English and Swedish, with strong written and verbal communication skills.
• Prior experience in AI red teaming, adversarial AI research, cybersecurity, penetration testing, or socio-technical probing.
• Demonstrated ability to systematically test complex systems for vulnerabilities, unexpected behaviors, and potential misuse.
• Strong understanding of structured testing approaches, including the use of taxonomies, benchmarks, frameworks, or established playbooks.
• Excellent analytical and critical-thinking skills, combined with creativity and curiosity when exploring unconventional attack paths.
• Strong ability to document technical findings clearly and explain risks to both technical and non-technical audiences.
• High attention to detail and commitment to producing reproducible, evidence-based evaluations.
• Comfortable working independently in a remote, asynchronous environment and adapting quickly across different projects.
• Experience with adversarial machine learning, cybersecurity, or socio-technical risk is preferred.
• Creative probing skills, including experience with psychology, acting, creative writing, or unconventional adversarial thinking, are an advantage.

Benefits:

• Competitive contract compensation of $48–$62 per hour.
• Fully remote work environment.
• Flexible, asynchronous working structure.
• Opportunity to contribute directly to the safety and responsible development of advanced AI systems.
• Exposure to diverse conversational AI models, agents, safety challenges, and adversarial testing methodologies.
• Opportunity to combine technical expertise with creative and unconventional problem-solving approaches.
• Work across varied AI safety projects and use cases while expanding expertise in adversarial testing.
• Meaningful contribution to improving AI model robustness, reliability, and responsible deployment.