I'm Mann, an independent Alignment Researcher investigating emergent misalignment through the lens of complex systems. My agenda is to understand the moment a model tips into misalignment as a critical transition, and to anticipate it before it happens rather than only noticing once it has. I ask whether that onset is gradual or abrupt, whether it can be reversed or locks in for good, and whether it grows sharper as models scale up.

My previous Research and Product Engineering background spanned Computer Vision for robotics and manfacturing, Autonomous Penetration Testing, Coding Agents and evaluation sandboxes for large enterprise systems , Scaling inference and performance of diffusion models, Methods for structured pruning and quantization, Applications of AI in healthcare and biomedicine.

I have also engaged with non-profits, consulting on the safe and ethical integration of AI and technology across conversational AI for educational ministries, analytics platform for oversight of grant allocation, and tools to combat racism and discrimination against minority groups.

For a lot of my work, I have received grants and have been a part of the Develop For Good, Digital Product School and PiSchool fellowships.

Alongside my research, I’m active in reading groups and EA India community. I co-founded SAGA (Safety and Alignment Group Ahmedabad) with Kulraj Chavda (FAR.AI), a local field effort to bring together researchers, policy makers, and VCs to drive actionable outcomes for the future of AI. We are tackling the talent bottleneck in technical AI safety, demonstrating and communicating importance of x-risk clearly to law makers and nudging investors towards formation of for-profit safety organisations.