Programme Thesis
The MATS Autumn 2026 Scholar - AI in Society fellowship funds independent research on AI alignment, security, and governance, providing mentorship from top researchers at organizations like Anthropic and Google DeepMind. It exists to accelerate early-career researchers toward producing tangible research outputs that address the societal risks and technical challenges of advanced AI systems.
Selection Criteria
- Research potential and clarity of independent research direction
- Relevance of proposed work to AI alignment, security, or governance
- Technical competence (e.g., machine learning, computational modeling, or relevant domain expertise)
- Ability to produce tangible research outputs within a 10-week intensive phase
- Fit with available mentors and labs (Institut Polytechnique de Paris, HEC Paris, Anthropic, Google DeepMind)
- Early-career stage and potential for long-term impact in AI safety
Past Winners / Cohort Profiles
Past MATS cohorts typically include early-career researchers (recent graduates, PhD students, or postdocs) with strong technical backgrounds in ML, neuroscience, or related fields. Archetypes include independent researchers with preprints or open-source projects, often from top universities or labs. Named examples are not provided on the page, but the program emphasizes mentorship from Anthropic and DeepMind researchers, suggesting winners have demonstrated ability to work on frontier AI safety problems.
Ideal Candidate Fingerprint
The platonic ideal applicant is an early-career researcher with a clear, technically grounded research proposal in AI alignment or governance, backed by demonstrable coding and modeling skills (e.g., Python, PyTorch, Bayesian methods). They have a track record of independent work (preprints, GitHub projects) and can articulate how their project will produce a concrete output (paper, tool, or analysis) within 10 weeks.
Recommended Framing
Eniola should frame their CCT model and neurocascade simulation as a direct contribution to AI safety: understanding reward-memory encoding in addiction offers a testbed for aligning AI reward systems and preventing pathological goal-seeking. Their multi-domain computational skills (ODE modeling, Bayesian calibration, protein ML) and independent preprint record demonstrate the technical independence and output orientation MATS seeks. Emphasize how their work on circuit-level pharmacology simulation can inform safe AI architectures, especially in reinforcement learning and value alignment.
Watch Out
The program is on-site in Boston/Cambridge, MA (US/Canada), which may pose visa or relocation challenges for a Nigerian applicant currently enrolled in Germany. The 10-week intensive timeline may conflict with M.Sc. coursework at HPI/Potsdam (Winter Semester 2026/27). No explicit mention of funding amount or stipend, which could be a concern for an independent researcher. The program's focus on AI alignment may require reframing neuroscience/pharmacology work to fit safety and governance themes.
Research History
2026-07-28 09:44 · medium confidence