Technical Fellowship
Some basic machine learning background is a plus, but not required.
9 weeksWeekly 2-hour cohortDinner providedHarvard SquareNext cohort: Spring 2027
Every semester and summer, AIST runs a nine-week introductory reading group on AI safety.
What you'll cover
- Expert predictions on AI progress, and when we can expect transformative impacts
- How frontier language models are trained, and how their safety can fail
- Why having the right training protocol might not be enough to guarantee the behavior we want from AI models
- Current research into the inner workings of neural networks
- The current plans to make human-and-beyond-level AI safe, and how effective they are on current AI systems
Questions
Who can join the fellowship?
Past intro fellows have primarily been undergraduate, masters and graduate students (as well as postdocs) from Harvard. We also occasionally accept students from other Boston-area universities. Non-student professionals are also welcome to apply, especially during the summer fellowship.
How much experience with machine learning is expected?
Participants should be familiar with basic concepts in machine learning such as deep neural networks, stochastic gradient descent and reinforcement learning. We may group cohorts according to previous experience. We also have a set of “Week 0” materials to introduce these topics.
Should I apply if I'm mostly interested in AI policy?
The Technical Fellowship presents AI safety from a technical perspective, often reading research papers on specific parts of the alignment problem. We don't discuss the governance implications, though this background is highly useful context for later policy work. The Policy Fellowship is likely a better fit for governance-oriented people.
How competitive is the application?
In Fall 2026, we received around 250 applications to the technical fellowship and accepted around half.
When and where does it meet?
We ask for your availability in the application and try to accommodate schedules when forming cohorts. Each cohort meets once a week for two hours, with dinner or lunch provided, in our office in Harvard Square.
What if I already know the material?
Email us at harvardaisafetyteam@gmail.com to talk about other ways of getting involved.
Who facilitates?
AIST members with research experience in AI safety, including upperclassmen and graduate students.
How often does it run?
We usually run Spring, Summer and Fall iterations.
Policy Fellowship
Some background with policy, government, or AI is a plus, but not required.
9 weeksWeekly 2-hour cohortDinner providedHarvard SquareNext cohort: Spring 2027
Every semester, AIST runs a nine-week introductory reading group on the foundational policy and governance issues posed by advanced AI systems.
What you'll cover
- How much progress in AI should we expect over the next few years?
- What are the risks associated with the misuse and misalignment of advanced AI systems?
- What are the tradeoffs between different theories of victory for AI governance, from pausing AI development to US-led AI dominance to cooperative international development?
- How could liability regimes and private governance mechanisms incentivize responsible AI development?
- How can the US reduce AI risks while navigating strategic competition with China?
- How can computing power serve as a lever for AI governance, and how effective have export controls been so far?
- What role can early interventions like mandatory lab disclosures and government AI capacity-building play as AI systems advance?
Questions
What sort of experience is expected?
Previous work in AI or technology policy is helpful but by no means required. Fellows have had backgrounds in public policy, law, political science and a variety of technical fields, and we encourage applicants from other backgrounds too.
How competitive is the application?
In Fall 2026, we received around 250 applications to the policy fellowship and accepted around half.
When and where does it meet?
Each cohort meets once a week for two hours, with dinner or lunch provided, in our office in Harvard Square. The summer fellowship has both in-person and virtual options.
Is it open to graduate and undergraduate students?
Yes. Previous fellows have included students in SEAS, GSAS and MIT PhD programs, HLS, HKS, HBS and Harvard College. Schedules permitting, we match students with others from similar backgrounds.
What if I already know the material?
Email us at harvardaisafetyteam@gmail.com to talk about other ways of getting involved.
AI × Biorisk Fellowship
Some background in AI or the life sciences is a plus, but not required.
9 weeksWeekly 2-hour cohortDinner providedHarvard SquareNext cohort: Spring 2027
AIST's newest nine-week introductory reading group, for anyone with a life-science background who wants to understand how AI is changing biorisk.
What you'll cover
- How AI can increase biological capabilities through design tools and experimental execution, including the creation of bioweapons
- Current gaps in safeguards, policies and benchmarks to diagnose and prevent AI-enabled biorisk
- How AI can also benefit biodefense: disease prevention, detection, screening and vaccine design
- How malicious actors can misuse AI tools and LLMs to create biosecurity risks
- Governance and policy solutions
- Technical safeguards: red teaming, evaluations and benchmarks
Questions
How much biology experience do I need?
The fellowship is intended for anyone with a life-science background at any level who is interested in how developments in AI are affecting biorisk.
How much experience with AI is expected?
None needed. Some background in AI, particularly its applications to biology, helps, but it's not a prerequisite.
How competitive is the application?
In Fall 2026, we received around 100 applications and accepted about half.
How often is it offered?
Fall 2026 is the first iteration. We plan to run it in Spring, Summer and Fall 2027 as well.
Membership
After completing one of the intro fellowships, Harvard students can apply for membership.
Members join a technical, policy or strategy track, get 24/7 access to our office in Harvard Square, and facilitate future fellowship cohorts.
Workshops
With MIT AI AlignmentFriday to Sunday3 times a semester
AIST and MIT AI Alignment (MAIA) run joint weekend workshops for members and fellows to meet professionals working on AI safety.
Each one mixes talks, small-group discussion and 1:1s. Researchers from Redwood Research, Anthropic, METR, OpenAI, RAND, CNAS, and HKS/HLS have joined past workshops.
Do I need to be a member to attend?
Attendees are generally members of AIST or MAIA, or participating intro fellows.
Technical or policy?
We run a mix of technical-focused and policy-focused workshops throughout the year.
When are they?
Over three weekends in the middle of the semester. Each workshop runs from Friday evening to Sunday afternoon.
Spring 2027 applications open soon
