Forward-Deployed AI Safety Researcher

Antler; United States; Hybrid; USD 29,000 - 44,000; Full time

Job summary

Antler is seeking an AI Safety Researcher to track evolving jailbreaks and exfiltration techniques for real model deployments. You will reproduce attacks, work with engineers to convert findings into detectors, and ship updates to customers during rapid cycles.

This forward-deployed role may require on-site work with customers during pilots and security reviews, with engagements lasting up to three weeks at a time.

Qualifications

• Strong background in ML/NLP, applied security research, or adversarial ML—academic or industry.
• Hands-on experience prompting, fine-tuning, or red-teaming LLMs.
• Able to read and reproduce findings from security research papers and disclosures.
• Able to communicate technical findings to engineers and non-technical stakeholders.
• Comfortable deploying on client sites during pilots and security reviews.

Responsibilities

• Research and reproduce emerging attack techniques against LLMs and autonomous agents (prompt injection, jailbreaks, data exfiltration, tool-call abuse).
• Design evaluation suites and red-team harnesses to test detectors against novel attacks.
• Translate research findings into production detector specs and policy primitives with engineering.
• Publish internal and external write-ups on notable findings.
• Map attack classes and detectors to compliance frameworks (NIST AI RMF, ISO 42001, EU AI Act).
• Advise customers on emerging risks during pilots and security reviews.

Skills

ML/NLP

Adversarial ML

Python

LLMs

Tools

Python tooling

Job description

Antler is seeking an AI Safety Researcher to track evolving jailbreaks and exfiltration techniques for real model deployments. You will reproduce attacks, work with engineers to convert findings into detectors, and ship updates to customers during rapid cycles.

This forward-deployed role may require on-site work with customers during pilots and security reviews, with engagements lasting up to three weeks at a time.