Research Engineer, Safety

Written by the T&S Network from Decagon's posting.

Safety research engineering for Decagon's conversational AI agents that handle enterprise customer support, covering evaluation through production. The work sits in the Research team, which builds the model and decision-making stack, and focuses on identifying real-world failure modes and building the models, evaluations, and safeguards that prevent them.

• AI misuse and jailbreaks

What you'd do

• Design safeguards that block prompt injection, tool misuse, and leaks of sensitive data
• Build safeguards that stop policy violations and invented commitments from agents
• Create stress tests, simulated runs, and red-team sets seeded by real production failures
• Ship classifiers, judges, reward signals, and runtime guards that shape agent behavior
• Analyze production traces and incidents to find root causes, test fixes, and measure impact
• Partner with security, legal, product, and customer-facing staff to turn enterprise needs into safeguards

What it asks for

• Two or more years in AI or ML engineering, research, or AI safety
• Hands-on time working with language models or agent systems, including evaluating, tuning, or deploying them
• Practice with reinforcement learning, preference tuning, distillation, or model routing
• Fluency in Python and modern ML tooling, with production engineering depth
• Exposure to human review loops, incident handling, or staged rollout practices for ML systems (preferred)

Terms in this role

Red teaming.

Testing a system, often an AI model, by deliberately trying to make it fail or misbehave, so problems are found and fixed first.

Our best summary of the posting, and it can be wrong. The employer's posting is what counts: read it in full on Decagonnorth_east before you apply. If we got something wrong, report this listing.