Marjia Siddik is the founder of Safentic, a runtime enforcement layer for AI agents. With a background in software engineering, AI safety, and policy, she’s focused on building systems that are both capable and accountable. Her work brings a practical approach to governing autonomous AI in the real world.
Selected Projects
A privacy-preserving data system that uses Path ORAM to hide access patterns by securely reshuffling data blocks.
A framework for mitigating bias in healthcare AI, with a datasheet template and schema for dataset documentation.
Work Experience
Research
Toward ASI Stability: A Treaty Framework for US–China Cooperation on Artificial Superintelligence
Supervisor: Corin Katzke
Organization: Convergence Analysis
This paper, developed through the Supervised Program for Alignment Research, proposes a phased treaty for US–China cooperation on artificial superintelligence. It combines arms control ideas with technical tools like compute tracking to reduce strategic risk. Set to be presented at Oxford’s 2025 workshop on equitable AI access.
Datasheets for Healthcare AI: A Framework for Transparency and Bias Mitigation
Supervisor: Dr. Harshvardhan J. Pandit
Organization: ADAPT Centre, Dublin City University
Conducted under the ADAPT Centre and funded by the Health Research Board, this research explores the use of datasheets to improve transparency and documentation in medical AI datasets. Published in CEUR Workshop Proceedings, it was presented at AICS 2024 (UCD), New Scientist Live UK 2024, and the ADAPT Scientific Conference 2025.
Blog
Connect
Feel free to contact me at marjiasdk@gmail.com