PeopleofSuperintelligence

Researcher

Chejian Xu

Chejian Xu is a final-year computer science Ph.D. candidate at UIUC who earned a B.S. in computer science from Zhejiang University; his research focuses on safe, scalable, and interpretable AI. He coauthored AdvGLUE and SafeBench, then coauthored DecodingTrust, a GPT trustworthiness assessment that received the 2023 NeurIPS Outstanding Paper Award. At NVIDIA Research and Meta Research, he worked on long-context language models and vision-language reasoning; his later papers tested web agents and multimodal models and extended LLM context to four million tokens. In 2026, he is a Student Researcher at Google DeepMind working on self-evolving agents.

Badges

Timeline

  1. 2021

    Adversarial GLUE

    At NeurIPS 2021, he coauthors Adversarial GLUE (AdvGLUE), a multi-task benchmark for evaluating language-model vulnerabilities to textual adversarial attacks.

  2. 2022

    SafeBench for driving safety

    At NeurIPS 2022, he coauthors SafeBench, a unified platform for evaluating autonomous-driving safety with varied scenarios, generation algorithms, routes, and environments.

  3. 2023

    Testing GPT trustworthiness

    He coauthors DecodingTrust, an assessment of GPT-4 and GPT-3.5 across toxicity, bias, robustness, privacy, machine ethics, and fairness. NeurIPS gives the paper its 2023 Outstanding Paper Award.

  4. 2024–2025

    Research internships at NVIDIA and Meta

    In 2024, he interns at NVIDIA Research on long-context LLMs. In 2025, he interns at Meta Research on VLM reasoning and interpretability.

  5. 2025

    Red-teaming agents and multimodal safety

    At ICML 2025, he coauthors AdvAgent, a black-box red-teaming framework with a reinforcement-learning pipeline for testing web agents. At ICLR 2025, he coauthors MMDT, a benchmark for multimodal models that assesses safety, hallucination, fairness, privacy, and robustness.

  6. 2026

    Long-context LLM training

    At Findings of ACL 2026, he coauthors an efficient training recipe that extends long-context LLMs from 128K to 4M tokens. Its UltraLong-8B model is built on Llama-3.1-Instruct.

  7. 2026

    Student Researcher at Google DeepMind

    In 2026, he works as a Student Researcher at Google DeepMind on self-evolving agents.

  8. As of October 2026

    Ph.D. candidate at UIUC

    He is a final-year computer science Ph.D. candidate at UIUC, advised by Prof. Bo Li, and holds a B.S. in computer science from Zhejiang University. His research focuses on safe, scalable, and interpretable AI.

Career

  1. 2026Google DeepMindStudent Researcher
  2. 2025MetaIntern
  3. 2024NVIDIAIntern

Education

  1. Zhejiang UniversityBS in computer science
  2. University of Illinois Urbana-ChampaignPhD candidate

From the biography.

  1. Google alumni have founded or led 133 other organizations, more than any other AI labIncludes Chejian Xu. We estimate that alumni of 17 major AI labs went on to found or lead 350 other organizations, including 226 they founded, and we trace 472 career moves directly between 23 labs since 2010.
  2. Google had the most people shaping AI every year from 2014 to 2025Includes Google. Our estimates of which organizations employed the most people shaping AI each year from 1955 to 2025, tracing the shift from university labs to companies.
  3. Google has the highest combined h-index among AI organizationsIncludes Google. We estimate that Google holds the highest combined h-index at 8,016, while universities account for 73% of the deduplicated total across AI organizations.
  4. 20% of people at companies who shape AI also hold a university role, led by Google and Stanford UniversityIncludes Google. We estimate that 539 of the 6,766 people with a known current organization (8%) hold both a university role and a company role, including 20% of people at companies and 13% of people at universities.

Colleagues

People named in this biography, and people with dated positions at the same organization in the same years.

Sources and credits

Photo: We have no published photo of Chejian Xu yet. Send us a free-licensed photo.

  1. Adversarial GLUE: A Multi-Task Benchmark for Robustness Evaluation of Language Models datasets-benchmarks-proceedings.neurips.cc. Checked 20 October 2026.
  2. SafeBench: A Benchmarking Platform for Safety Evaluation of Autonomous Vehicles proceedings.neurips.cc. Checked 20 October 2026.
  3. DecodingTrust: A Comprehensive Assessment of Trustworthiness in GPT Models arxiv.org. Checked 20 October 2026.
  4. DecodingTrust: A Comprehensive Assessment of Trustworthiness in GPT Models - Microsoft Research microsoft.com. Checked 20 October 2026.
  5. Chejian Xu xuchejian.com. Checked 20 October 2026.
  6. AdvAgent: Controllable Blackbox Red-teaming on Web Agents proceedings.mlr.press. Checked 20 October 2026.
  7. MMDT: Decoding the Trustworthiness and Safety of Multimodal Foundation Models - Microsoft Research microsoft.com. Checked 20 October 2026.
  8. From 128K to 4M: Efficient Training of Ultra-Long Context Large Language Models - ACL Anthology aclanthology.org. Checked 20 October 2026.