AI Jailbreak And Prompt-Injection Security Expert

The Work

You will assess how AI systems respond to adversarial prompts and security threats. The work combines hands-on testing with clear reports and recommendations for improving model safety.

You will create repeatable evaluation methods, maintain regression tests, and explain technical findings to both technical and non-technical audiences.

• Design tests for jailbreaks, prompt injection, tool-use abuse, and other unsafe model behaviors.
• Develop multi-turn and cross-domain strategies that uncover complex bypass patterns.
• Build evaluation frameworks that reflect realistic adversarial threats.
• Maintain and update regression test suites for jailbreak and prompt-injection weaknesses.
• Turn security findings into practical model-safety and risk-reduction recommendations.
• Document methods, findings, and best practices in reports and presentations.

What It Pays And Takes

• Pay: $50 to $90 per hour.
• Work type: Part-time contract work.
• Location: Open to candidates in the countries listed for this role, including the United States, Canada, India, and other listed countries.
• Language: Professional fluency in English.
• Experience level: Entry level.
• Required: Experience researching, testing, or identifying jailbreak, prompt-injection, tool-use abuse, or other adversarial AI vulnerabilities.
• Useful knowledge: Large language model architectures, prompt engineering, and security assessment methods.
• Helpful background: An advanced degree in computer science, cybersecurity, machine learning, or a related field, or equivalent professional or operational experience.
• Helpful evidence: Published research, open-source security tools, conference presentations, or work in multidisciplinary AI safety initiatives.
• Strong written and verbal communication skills are required so you can explain technical findings clearly.

How It Works

Apply on OpenTrain with your resume and then complete the application on the hiring site.

About AI Training Work

AI training is the human work behind systems that learn from examples, including testing model responses and reviewing safety risks. People with specialized security and machine learning knowledge are needed to find weaknesses and help make these systems more reliable.