AI Security Expert

The work

Probe AI systems for security weaknesses and unsafe behavior, then help engineering teams improve their defenses. This remote hourly contract involves 10 to 40 hours per week through independent, asynchronous assignments.

Responsibilities

• Run adversarial security exercises to find weaknesses and behaviors that can be exploited.
• Create prompts and unusual scenarios that test the limits of model safeguards.
• Assess generated responses for safety problems, bias, and failures to meet policy requirements.
• Record vulnerabilities, exploitation attempts, and unexpected outcomes in organized reports.
• Recommend practical mitigations with engineering colleagues.
• Follow emerging threats, safeguard bypass methods, and changes in security practice.
• Contribute to the criteria and procedures used for security evaluation.

Background and skills

Applicants need a strong grounding in cybersecurity, such as threat modeling, penetration testing, or vulnerability research. Practical experience with AI or machine learning systems, large language models, or prompt engineering is required.

You should approach investigation creatively and analytically, write clear documentation for technical readers, and manage task based work without close supervision.

Additional strengths

An information security, ethical hacking, or AI safety research background is helpful. Familiarity with open source AI tools or evaluation frameworks is an additional advantage.

Working arrangements

This is a remote hourly contract with 10 to 40 hours per week. Work is organized around independent assignments and a flexible schedule. The work involves collaboration with security and AI researchers.

Strong contributions may lead to ongoing assignments, broader responsibilities, or contract extensions.