University of Massachusetts Amherst
The University of Massachusetts Amherst hosts the SCALAR (Safe, Confident, and Aligned Learning + Robotics) Lab, directed by Associate Professor Scott Niekum in the Manning College of Information and Computer Sciences. Niekum's research aims to ensure AI systems are well-aligned with human objectives and can be deployed safely in the real world, developing algorithms that enforce safety constraints, provide performance guarantees, and infer human intentions. His work spans imitation learning, reinforcement learning, reward inference, and AI safety in contexts ranging from large language models to robotics. UMass CICS also hosts a broader AI Safety Initiative co-led by Professors Shlomo Zilberstein and Eugene Bagdasarian, advancing research on multi-agent AI safety, AI alignment, robustness, interpretability, and governance.
Funding Details
- Annual Budget
- -
- Monthly Burn Rate
- -
- Current Runway
- -
- Funding Goal
- -
- Funding Raised to Date
- -
- Fiscal Sponsor
- -
Theory of Change
Niekum and the SCALAR Lab believe that many AI alignment and safety issues are best addressed at the design stage rather than through post-hoc fixes. By developing technical methods for reward inference, safe reinforcement learning, probabilistic performance guarantees, and agent alignment verification, they aim to build AI systems that are structurally safer before deployment. Their work on emergent misalignment and rare event estimation in LLMs addresses the risk that AI systems may behave in unexpectedly harmful ways even when trained to appear aligned. The underlying theory is that rigorous technical foundations for alignment — grounded in human preferences and downstream outcomes rather than proxies — are necessary to prevent catastrophic outcomes as AI systems become more capable.
Grants Received
from Open Philanthropy
from Open Philanthropy
from Long-Term Future Fund
Projects
No linked projects.
People
No linked people.
Discussion
Sign in to join the discussion.
No comments yet. Be the first to share your thoughts.
Details
- Last Updated
- Apr 2, 2026, 9:54 PM UTC
- Created
- Mar 19, 2026, 10:42 PM UTC