Summary

The Harvard AI Safety group functions as a community of technical and policy researchers dedicated to mitigating risks within artificial intelligence and steering its future trajectory toward a beneficial direction. Our primary activity involves organizing a semester-long introductory reading group focused specifically on AI safety research, which aims to educate attendees on critical issues such as neural network interpretability, how models learn from human feedback, the risks associated with goal misgeneralization in reinforcement learning agents, the extraction of latent knowledge, and the evaluation of dangerous capabilities in advanced models. Through these structured sessions, the community shares practical perspectives and deep technical insights to foster a robust environment for future AI development. This approach ensures that emerging challenges like adversarial attacks and catastrophic forgetting are proactively addressed, allowing participants to contribute thoughtfully and effectively to the field. The group also serves as a bridge between researchers, educators, and practitioners, creating a collaborative ecosystem where theoretical concerns are validated through real-world scenarios. By focusing on these specific domains, the community ensures that safety considerations remain central to every aspect of AI engineering and policy, promoting responsible innovation and secure human-AI interaction.

Title

AISST

Description

AISST

Keywords

technical, policy, fellowship, resources, papers, research, safety, menu, home, team, mission, workshops, intro, folder, back, risks, open

NS Lookup

A 198.49.23.144, A 198.49.23.145, A 198.185.159.145, A 198.185.159.144

Dates

Created 2026-04-13

Updated 2026-05-01

Summarized 2026-05-01