AI Safety Expert for English and Swedish Red Teaming

Act as a human red teamer who probes conversational AI systems in English and Swedish to uncover safety vulnerabilities, produce high-quality adversarial inputs, and deliver reproducible artifacts customers can act on. The work is fully text-based, focuses on sensitive topics such as bias, misinformation, and harmful behaviors, and includes optional higher-sensitivity engagements supported by clear guidelines and wellness resources.

• Perform adversarial testing of conversational AI models and agents, including jailbreaks, prompt injection, misuse scenarios, bias exploitation, and multi-turn manipulation.
• Generate human-derived safety data: annotate model failures, classify types of vulnerabilities, and flag systemic risks.
• Follow established structure: use taxonomies, benchmarks, and playbooks to ensure consistent, comparable testing.
• Document findings reproducibly: produce reports, datasets, and attack cases that customers can reproduce and act on.
• Optionally participate in higher-sensitivity reviews, with topics communicated in advance and access to wellness resources when applicable.

• Native fluency in both English and Swedish, spoken and written, is required.
• Prior red teaming experience, such as AI adversarial work, cybersecurity testing, or socio-technical probing.
• A curious, adversarial mindset with a habit of pushing systems to breaking points while working methodically.
• Ability to communicate risks clearly to both technical and non-technical stakeholders.
• Comfort working across projects and adapting to different customers and requirements.

Preferred specialties

• Adversarial ML techniques, for example jailbreak datasets, prompt injection, RLHF or DPO attacks, and model extraction.
• Cybersecurity experience, such as penetration testing, exploit development, or reverse engineering.
• Socio-technical risk work, including harassment, disinformation probing, abuse analysis, and conversational AI testing.
• Creative probing skills from psychology, acting, or writing to design unconventional adversarial approaches.

• Remote engagement, text-only work.
• This is an hourly engagement, paid at the rates listed below.
• Participation in higher-sensitivity projects is optional, topics are disclosed before exposure, and wellness guidance and other support are provided for sensitive work.

• 48 - 62 hourly

• Native fluency in English and Swedish is mandatory.
• Ability to work remotely and to review AI outputs that may involve bias, misinformation, or harmful behaviors. Participation in sensitive-content reviews is voluntary and will be clearly communicated in advance.