Carlos Guerrero

I'm a Dartmouth alumnus (BA in CS & Math, MSc in CS) working on AI alignment. I'm a research fellow at CBAI, a research contractor at Redwood Research, and a SPAR fellow, and I previously worked with Soroush Vosoughi at Dartmouth on deep learning and NLP, with published work in NAACL 2025, ICML 2026, and NeurIPS 2026 workshops.

Experience

• Research Fellow — CBAI, Boston, MAOct 2026 – Present
  Working on alignment research with Professor Shi Feng.
• Research Contractor — Redwood Research, RemoteSep 2026 – Present
  Working on various alignment evaluations for the GPT-6 family, both Fable models, and Opus 5.5, as well as multiple open-source models.
• SPAR Fellow — SPAR, RemoteSep 2026 – Present
  Working on the project Moral Shadowbanning: Is Conversational AI Less Helpful, Honest, or Harmless to Users Deemed to Have Poor Moral Character? with mentors Pranav Mahajan and Lydia Nottingham.
• Researcher — Minds, Machines and Society Group, DartmouthJun 2024 – Present
  Worked with Professor Soroush Vosoughi on my undergraduate and Master's theses, and on multiple projects, some currently under peer review. Published papers in NAACL 2025, ICML 2026, and NeurIPS 2026 workshops, with the NeurIPS paper as first author.

Publications

• Beyond Watching: Active Behavioral Monitoring of Black-Box AI Agents
  Nalagatla, G., & Guerrero Alvarez, C.
  NeurIPS 2026 AIWILD Workshop
• PolicyLLM: Neuro-Symbolic Policy Extraction and Enforcement for Runtime AI Governance
  Haider, S. A., Huh, B., Kim, H. H., Nalagatla, G., Guerrero Alvarez, C., Raj, Y., & Vosoughi, S.
  ICML 2026 Workshop on Technical AI Governance Research
• Making VLMs Less Blind: A Neuro-Symbolic Approach
  Guerrero Alvarez, C.
  Master's Thesis, Dartmouth College, 2026
• 1 Layer Induction Heads and Some Research
  Guerrero, C., & Nalagatla, G.
  LessWrong, 2026
• What is it? Towards a Generalizable Native American Language Identification System
  Yang, I., Ma, W., Guerrero Alvarez, C., Dinauer, W., & Vosoughi, S.
  NAACL 2025 Student Research Workshop
• Evaluating Interoperability of Medical Communication Protocols via Differential Testing
  Anantharaman, P., Varadharaju, V., Guerrero Alvarez, C., Borodchuk, D., Beg, A. A., Shapiro, R., Gettinger, A., Smith, S. W., & Locasto, M. E.
  Cybersecurity in Healthcare

Selected Projects

• Tilpa Dev — An SDK for building real-time adaptive APIs.
• Stockformer — A transformer-based trading strategy with thorough backtesting.
• MicroThink — Small LLM made with Mamba blocks instead of Transformers.

Last updated October 2026