AI Red Team Researcher

Who we are.

Reality Defender is an award-winning cybersecurity company helping enterprises and governments detect deepfakes and AI-generated media. Utilizing a patented multi-model approach, Reality Defender is robust against the bleeding edge of generative platforms producing video, audio, imagery, and text media. Reality Defender's API-first deepfake detection platform empowers teams and developers alike to identify fraud, disinformation campaigns, and harmful deepfakes in real time.

Backed by world class investors including DCVC, Illuminate Financial, Y Combinator, Booz Allen Hamilton, IBM, Accenture, Rackhouse, and Argon VC, Reality Defender works with leading enterprise clients, financial institutions, and governments in order to ensure AI-generated media is not used for malicious purposes.

Youtube:
Reality Defender Wins RSA Most Innovative Startup

Reality Defender's AI Red Team exists to break our own detection models before adversaries do. As a Red Team Researcher, you'll get your hands on the newest deepfake tools the moment they appear — voice clones, real-time face swaps, injection attacks, multimodal impersonations — use them the way real attackers would, and turn every failure you uncover into a measurable fix shipped alongside our AI and Data teams.

This is a full-time role based at our NYC headquarters or remote within the U.S., reporting to VP of Product.

Responsibilities include:

• Run continuous adversarial testing against our detection models across audio, video, image, and text, reproducing new generative tools and techniques as they're released and stress-testing them against production models.

• Build realistic, end-to-end attack simulations that mirror how bad actors actually operate — cloned voices in call centers, real-time synthetic video in onboarding flows and job interviews, virtual camera and media injection, and coordinated cross-channel impersonation.

• Build and maintain repeatable red team infrastructure: test harnesses, attack catalogs, and an evolving adversarial dataset that every model release is validated against.

• Close the loop on findings — prioritize them, hand them to AI scientists and engineers with clean reproduction steps, and verify fixes through regression testing so they stick.

• Track the generative AI landscape (open-source models, commercial tools, research papers, and threat actor tradecraft) and triage which techniques are most likely to be weaponized next.

• Partner with Product to shape roadmap priorities based on what red team findings reveal.

• Support Sales, Marketing, and Solutions Engineering with credible, publishable research and threat briefings for enterprise and government customers.

• Contribute to adversarial testing of the broader platform — APIs and integrations — in partnership with Security.

• Help define the ethical guidelines and safe-handling practices for synthetic media used in testing.

• 4+ years of experience in offensive security, adversarial ML, AI red teaming, or ML engineering with a strong adversarial focus.

• Hands-on experience generating synthetic media — voice cloning and TTS, face swap, lip-sync, diffusion-based image and video tools — in an offensive or research context, not just reading about it.

• Strong Python skills and practical ML fluency (PyTorch or similar): you can build attack tooling, automate testing pipelines, and reason about why a model fails, not just that it did.

• Working knowledge of adversarial ML techniques — evasion, adversarial perturbations, compression and post-processing attacks, data poisoning — and how to measure model robustness.

• Clear communicator who can turn technical findings into prioritized, actionable guidance for engineers and non-technical audiences alike, and who…