Bring your red teaming and adversarial security skills to remote contract work on model safety. You find jailbreak and prompt-injection weaknesses in language models and build the test suites that keep them fixed. Your security knowledge is what counts here.
What you will do
- Design and run methods to check model safety, with a focus on ethical jailbreaks, LLM red teaming, prompt injection and tool-use abuse.
- Build cross-domain strategies that expose multi-turn and complex adversarial bypass patterns.
- Build and maintain regression test suites for jailbreak and prompt-injection weaknesses.
- Build evaluation frameworks that stress-test models against real-world adversarial threats.
- Work with technical stakeholders to turn security findings into fixes and risk reduction.
- Write clear reports and presentations on methods and findings for technical and non-technical readers.
What you need
- 2+ years in adversarial machine learning, LLM red teaming, model safety evaluation or a closely related security field.
- Proven experience finding weaknesses such as jailbreaks, prompt injection, tool-use abuse or adversarial attacks.
- An advanced degree (PhD or MS) in computer science, cybersecurity, machine learning or a related field, or equivalent professional background.
- Recognition in the security or adversarial ML community, such as published research, open-source tools or conference talks.
- Excellent written and spoken communication, with a focus on clear documentation.
- Familiarity with current LLM architectures, prompt engineering and security assessment tools.
The listing names these as the qualifications it would like to see. Earlier work on cross-functional safety projects is a plus.
How pay works
This job is paid hourly. The listed rate is $50-90/hr.
Apply (opens in a new tab)