About The Position

NVIDIA is seeking a Deep Learning Scientist to join its AI Security and Responsibility efforts. The role advances security alignment for large language models and develops guardrails to help AI systems resist adversarial manipulation and operate within trusted boundaries. The work focuses on securing multimodal agentic reasoning systems and carrying research from experiments into practical solutions. The scientist will apply security research to develop, train, fine-tune, and deploy models and guardrails across text, audio, images, video, and tool use. The role also advances post-training and alignment approaches, researches security techniques beyond post-training, and develops and benchmarks multimodal security datasets. The scientist will collaborate with engineering, data science, research, and product teams across NVIDIA. The expected base salary ranges are $187,500–$364,000 for Level 4 and $232,500–$442,000 for Level 5. The position is also eligible for equity and benefits. Applications will be accepted at least until November 22, 2026. This posting is for an existing vacancy. NVIDIA uses AI tools in its recruiting processes and is an equal opportunity employer.

Requirements

  • Master’s degree or equivalent experience, or PhD, in Computer Science, AI, or Applied Math.
  • At least 8 years of relevant work experience.
  • Excellent Python programming skills and strong fundamentals in scalable model development and deep learning frameworks such as PyTorch.
  • Strong knowledge of machine learning and deep learning techniques and modern foundation model architectures, including Transformers and mixture-of-experts models.
  • At least 1 year of dedicated experience or research in one or more of: LLM security (including backdoors, data poisoning, latent malicious behavior, or structural model vulnerabilities); frontier risks (including model deception, manipulation, or loss-of-control scenarios); agentic safety (including multi-turn tool-calling, orchestration, or execution risks); or multi-turn security evaluation using robust, scalable automated evaluation for conversational and iterative use cases.
  • Hands-on experience post-training multimodal models for multimodal reasoning.

Nice To Haves

  • Strong publication record at top-tier AI and machine learning venues, such as NeurIPS, ICML, ICLR, CVPR, LLMSEC, OMMM, or WAITI.
  • Active contributions to open-source AI safety tools, benchmarks, datasets, or models.
  • Experience with multilingual AI security.
  • Experience with automated red-teaming.

Responsibilities

  • Apply fundamental and applied security research to develop, train, fine-tune, and deploy large language models for multimodal agentic reasoning systems and guardrails covering text, audio, images, video, and tool use.
  • Advance post-training and alignment methods, including instruction tuning, preference optimization, and RLHF/RLVR/MOPD, to improve security of multimodal agents in complex use cases.
  • Research and deploy model and system security techniques beyond post-training, such as instruction hierarchy, risk detection, and new cyber gyms.
  • Lead the collection and development of multimodal security datasets and benchmark them to evaluate model accuracy, safety, and task-completion success.
  • Partner with engineers, data scientists, research teams, and product managers across NVIDIA to scale LLM security and agentic safety solutions.

Benefits

  • Comprehensive benefits package
  • Equity
© 2026 Teal Labs, Inc
Privacy PolicyTerms of Service