> Markdown version of [/jobs/ext/4577209-ai-red-team-engineer](https://www.wearedevelopers.com/jobs/ext/4577209-ai-red-team-engineer). Every page supports `.md` or `Accept: text/markdown`. Links point to the HTML versions so they work for humans too. Agent guide: [/agents.md](https://www.wearedevelopers.com/agents.md). --- # AI Red Team Engineer - **Company:** HUMMINGBIRDS AI INC. - **Location:** Austin, TX, United States - **Experience:** Experienced - **Contract:** Permanent contract - **Skills:** Artificial Intelligence, Fuzz Testing, Python (Programming Language), Regression Testing, Red Team (Cyber Security), TypeScript, Backend, Agentic-AI, Evaluation of Large Language Models - **Published:** November 5, 2026 - **Apply:** https://www.indeed.com/viewjob?jk=cb194734afffc6f1 ## About the Role * 3+ years building backend services, async workers, or test infrastructure in Python or TypeScript, with the failure cases and the tests. * An adversarial mind: offensive tooling, CTFs, fuzzing, or a security project. The title does not matter. * Judgment about evidence: when an automated evaluator is enough and when a human has to look. * Idempotency, backpressure, observability, and limits that are enforced rather than documented. Nice to have * Agent frameworks, voice or telephony systems, or LLM evaluation infrastructure * A security finding of your own, responsibly disclosed ## Description Our researchers find the failure by hand. You make it happen a thousand times, against authorized targets, with evidence a customer can inspect, and again on every release. You own the harness: the runner that places calls and chat sessions, the evaluators that decide what a result means, and the replay that shows what changed. Thousands of concurrent calls, bounded, cancellable, observable., * Generate useful variations. Branch multi-turn scenarios and voice conditions with the objective and conditions explicit. * Make results trustworthy. A successful attack, a refusal, an infrastructure failure, and an inconclusive result are distinguishable and inspectable. * Make findings reproducible. Scenario, configuration, model versions, tool events, outcomes, replay, and comparison. * Keep fixes fixed. Validated findings become regression tests that run on every change, with noise controlled.