AI Alignment Foundation Fellowship 2026 Fully Funded

Applications are now open for the AI Alignment Foundation Fellowship 2026, an 8-week, fully funded, full-time technical research program in AI Alignment. This prestigious fellowship offers a $12,000 stipend and is designed for talented researchers who want to work on real, high-impact AI alignment problems alongside experienced researchers.

The program runs from September 8 to October 30, 2026 and is fully remote. It is fully funded with no application fee. Selected fellows receive professional research support, compute resources, and career guidance..You can check out BS, MS, PhD and postdoc positions here.

Key Details:

Fellowship Name: AI Alignment Foundation Fellowship 2026
Host Organization: AI Alignment Foundation (in partnership with AE Studio)
Host Country: Remote (United States based organization)
Duration: 8 weeks (full-time, 40 hours per week)
Application Deadline: August 17, 2026 (Anywhere on Earth)
Application Fee: None
Open for: All nationalities
Funding Type: Fully Funded

Interested in More Positions? Follow links below

What This Fellowship Offers:

Participants receive comprehensive support that allows them to focus entirely on impactful research:

$12,000 stipend paid in installments for the full 8-week program
Full compute budget including cloud GPU access and API credits (no personal cost for experiments)
Dedicated research manager for daily support and project management
Guidance from AE Studio researchers and senior leadership through midpoint check-ins
Professional research infrastructure (Linear, GitHub, Slack, experiment standards)
Opportunity to produce a LessWrong post and potentially a peer-reviewed paper
Hybrid Demo Day presentation in Los Angeles with travel and accommodation fully covered (remote option also available)
Strong career support and networking with leading organizations in AI alignment
Valuable research output and portfolio pieces for future academic or industry opportunities

Also Check: PhD, Postdoc & Professor positions at the University of Freiburg, Germany.

Who Should Apply:

Strong candidates have:

Strong ML background — you can run experiments, implement papers, and debug your own code.
Spent real time thinking deeply about AI alignment.
Some form of independent research experience (LessWrong post, hackathon project, thesis, preprint, etc.).
Ability to work autonomously and show meaningful progress.
Openness to any career stage — recent PhDs, MSc students, self-taught researchers, or career transitioners are welcome.

This fellowship is probably not for you if:

You are primarily interested in policy or governance (this is a technical research fellowship).
You need significant hand-holding to start on new problems.
You have never written code to run experiments.
You are applying only for the credential and not genuinely interested in alignment long-term.
You have a fixed personal research agenda (projects are assigned based on fit).

Also Check: Technical University of Denmark PhD and Postdoc Scholarships 2026 (Fully Funded)

Program Structure:

You will receive dedicated support throughout the program:

Your Research Manager Provides day-to-day support, weekly prioritization, experiment tracking, and keeps the project moving.
AE Researchers Set the research vision and provide direction and feedback.
Midpoint Check-in One-on-one session with senior leadership for research guidance and career advice.

Every team works toward a LessWrong post (and potentially a paper). The program ends with a Demo Day presentation.

Also Check: Deakin University Postgraduate Research Scholarship (DUPR) Australia

Current Research Projects at AI Alignment Foundation:

The AI Alignment Foundation is actively working on the following high-impact research projects in AI alignment and technical safety:

1. Modular Pretraining Enables Access Control

Description: Explores how modular pretraining methods can create more controllable and secure AI systems by limiting knowledge access.

2. Self-Interpretation in Language Models via Adapter Probes

Description: Investigates techniques that allow language models to interpret and explain their own internal representations using adapter probes.

3. Steering Resistance: Self-Correction Circuits in Large Language Models

Description: Studies self-correction circuits that help models resist harmful steering and maintain alignment properties.

4. Self-Referential Processing and Introspection Across LLM Architectures

Description: Examines how different large language model architectures handle self-referential processing and introspection capabilities.

5. Neural Empathy as an Alignment Technique

Description: Researches whether neural empathy mechanisms can be used as a practical alignment technique to reduce harmful behavior in AI systems.

6. Self-Modeling in Neural Networks: Emergent Simplification and Robustness

Description: Explores how neural networks develop self-modeling capabilities and how this leads to emergent simplification and improved robustness.

7. Reward Model Discrepancies: Fine-Tuning Creates a False Sense of Safety

Description: Analyzes the gaps between reward models and actual model behavior during fine-tuning, highlighting risks of false safety signals.