Software Engineer, Trust & Safety

OpenRouter is an AI routing and infrastructure platform that helps enterprises access, manage, and optimize large language models across providers. The Software Engineer will build and operate trust and safety systems for detecting, investigating, and enforcing against abuse, fraud, malicious users, and illegal content across inference, payments, review tooling, and data platforms.

Responsibilities:

• Build and operate technical systems across signup, payments, and usage to detect abuse and fraud early
• Own the technical enforcement pipeline end to end, from detection proposing a candidate to a human reviewing it to restrictions landing across our systems
• Build the internal tools that support investigation and enforcement decision-making. This involves building case queues, evidence summaries, bulk review and enactment, and alerting mechanisms to trigger investigations
• Ship technical solutions for content safety on the inference path: illegal content detection and reporting, the KYC systems or external intelligence sources to allow us to more efficiently and proactively detect and stop fraud and abuse
• Help make abuse uneconomical by building systems, tools, and heuristics for quick detection and scalable enforcement
• Help investigate incidents directly in the data and build the analytics plane so we can quickly establish what happened, size the pattern, and separate real abuse from a false-positive cluster
• Build defensible monitoring for risk, abuse spikes, and fraud, all while reducing false positives
• Set the technical direction for how we prevent abuse as the platform grows and evolves, and define the patterns other engineers build against
• Work with data scientists on feature exploration and training of risk and abuse ML models

Requirements:

• 4+ years building and operating production systems, ideally including work in trust & safety, fraud, payments risk, security, or anti-abuse
• Proficient in React, TypeScript, Next.js, and JS runtimes
• Ability to write SQL against large event datasets and reason about base rates, precision and recall, and the cost of a wrong decision
• Sound judgment when working with incomplete evidence
• High agency and a bias toward action; spotting problems, iterating, and shipping without waiting for tickets
• AI-forward in your workflow, using coding agents often, finding interesting ways to automate your work, and holding opinions about what works and what doesn't
• Comfortable in a small, fast-moving environment where boundaries between teams are intentionally fluid
• Discretion and resilience, since parts of this role may involve reviewing or discussing disturbing content and handling sensitive user data
• Strong written and verbal communication, able to explain incident findings, detection rules, and systems to a range of audiences
• Motivated by adversarial problems, where the counterparty adapts to whatever you ship and the work keeps evolving
• Experience with payments fraud tooling (Stripe Radar, chargeback and dispute flows, crypto payment risk), or with identity and KYC systems
• Experience with LLM-specific abuse: jailbreaks, prompt injection, key theft and resale, shared or scraped credentials, automated account farming
• Experience operating large scale analytical datastores (e.g. ClickHouse, BigQuery) and observability platforms
• Familiarity with the reporting and compliance obligations that come with hosted AI, including illegal content reporting and model provider policy requirements
• Existing user of OpenRouter, or active side projects in AI products/infrastructure or developer tooling