Project summary

As the first AI Safety community in Taiwan, NTU AI Safety has run a reading group for three years. This year, we are scaling our community from 40 to 500+ people. We aim to bridge the gap in AI Safety between Taiwan and the global community. We run reading groups to discuss technical AI Safety techniques and the latest research papers, and we invite professional speakers from organizations such as METR and Anthropic, as well as upcoming AI Safety PhD researchers. We also actively seek partnerships and collaboration opportunities with global AI Safety organizations.

Taiwan has a strong pool of AI research talent and plays an important role in the global AI ecosystem, whether in technical AI research, biomedicine, or hardware. However, there are currently limited formal pathways for people who want to contribute to AI Safety, and not many people are familiar with the field. As the first AI Safety study group in Taiwan, we believe that expanding our community and reaching not only students but also early- and mid-career people is critical to building a stronger pipeline of talent for AI Safety. To ensure the sustainability and continued growth of our organization, we are seeking funding to expand our programs and strengthen our organizational capacity.

What are this project's goals? How will you achieve them?

We aim to help more people understand what AI Safety is, how they can contribute, what potential opportunities are, and how their existing professional expertise can be applied to preventing and mitigating AI risks. We have observed that while many people are increasingly aware of the potential risks of AI, they often don’t know where to start or how they can contribute.

We think having an AI Safety community can be an open door for people interested in AI Safety from different backgrounds by helping individuals to be more informed about opportunities and the talent pipeline. It’s also a nice opportunity to build connections with similar values.

Last semester, we organized 2 reading group tracks (Fundamentals and Technical Paper Reading) and 3 speaker events. We had 20 active members (students and working professionals) participating in our reading group and 50 people joining our speaker events. We also established an online presence through Discord, LinkedIn, and our website. We are planning to rerun both tracks.

This semester, we are organizing a 3rd reading group track on AI policy and governance. However, we recognized that there are still relatively few people working on AI policy and governance in Taiwan, after discussing with some members in our reading group with a law background, and we had limited international collaboration or perspectives to bring into the community.

In addition, there were only four members preparing for the weekly reading group last semester. The workload was heavy and stretched our capacity to the limit.

Therefore, our actions this year

• Expand the reading group team: recruit 7 facilitators to share the workload, rotate weekly facilitation, and make the program more sustainable as participation grows.
• Expand reading group track: Add AI Governance & Policy track to target law, philosophy, politics, and social science professionals.
• Reach new communities: Extend outreach to people from different professional and cause areas who may have relevant expertise or an interest in AI Safety.
• Build international connections:
  • Build an international reading group to connect with universities abroad and explore opportunities for in-person visits and exchange. Potential collaborators include MIT, Oxford, and UChicago.
  • Collaborate with AI Safety communities across Asia to develop an AI 2040 for APAC roadmap.
• Organize speaker events: Organize 3–4 speaker events on underexplored AI Safety topics, including international speakers in hardware verification and AI policy, either online or in person.
• Build cross-sector connection dinner: Organize connection dinners that bring together students and community members with professors and industry professionals
• Establish additional social media channels: Create and actively promote our Instagram, Facebook, and Luma channels to expand our outreach and reach people beyond the existing AI Safety community.
• Start preparing to support other university groups: To accelerate the field-building progress, we will also try to find other high-ranking university students and provide mentorship for establishing more university groups.

Potential future initiatives

To leverage the advantage of the hardware industry in Taiwan, we are also exploring a Hardware Verification track in future semesters to attract more hardware engineer, researchers, and students to join.

How will this funding be used?

The funding will cover organizers’ stipends and program staff (part-time) for 13 weeks. (FTE = 40hr/week)

Note: The reading group will run for 10 weeks in the semester (we ran it weekly) and 3 weeks of buffer for preparation

Core Organizers: Lead the overall development and operations of NTU AI Safety, including program planning, fundraising, partnerships, events, communications, and logistics. This role requires 0.8 FTE at a rate of $30/hour, totaling $12,480 over 13 weeks.

Facilitators: Support and deliver community programs, including curriculum preparation, session facilitation, member coordination, communications, and related logistics. This role requires 1.1 FTE at a rate of $30/hour, totaling $17,160 over 13 weeks.

Subtotal: 1.9 FTE, totaling $29,640 over 13 weeks.

10% Buffer: $2,964 is reserved for unexpected workload, operational needs, and potential payment or administrative fees, including a fiscal sponsorship fee.

Total Budget: 1.9 FTE, totaling $32,604 over 13 weeks.

Who is on your team? What's your track record on similar projects?

Team

Core Organizers

The core organizers have been building NTU AI Safety from scratch as volunteers since last semester.

• Chia-Tai (Teddy) Chang — Kairos Pathfinder Fellow, 2026, Bachelor's student in Computer Science at NTU
• Tzu-Tsen (Christine) Hsieh — Independent, currently supports NTU AIS full-time |  Master's graduate from NTU Electrical Engineering

Communications

• Ting-Yu (Amice) Lin — SASH Visitor

Track Organizers

• **Fundamental Track: **
  • Meng-Heng (Jasper) Tsai — Bachelor's student in Computer Science and Information Engineering, NTU
  • Tang-Yi (Jonathan) Wang — Master's student at the Graduate Institute of Electronics Engineering, NTU
  • Orina Shiue — LLM Engineer in TAIDE
• Technical Track
  • Yen-Shan (Lily) Chen — Master's student in Computer Science and Information Engineering, NTU
  • Leonardo Haw-Yang Foo — Master's student at NTU's Graduate Institute of Communication Engineering
  • Hao-Tian (Ted) Lu — AI Research Engineer
  • Yu-Chen Tsai — Bachelor's student in Math at NTU and currently a researcher at the HCI lab.
• **Policy Track: **
  • David Sánchez García — AI Governance Researcher
  • Leslie — Judicial Reform Foundation
  • Samanvya Hooda — Institute of Chinese Studies, Delhi, Bluedot AGI strategy facilitator

Track record (from last cohort)

• Community growth: Our Discord community (launched in February) has grown from 20 members last year to 400+ members today.
• Fundamentals & Policy Weekly Reading Group: An 8-week reading group based on the BlueDot Technical AI Safety course, supplemented with AI governance and policy articles, with around 13 stable, active attendees.
• Technical Paper Weekly Reading Group: An 8-week reading group covering a range of AI safety research papers, with around 14 stable, active attendees.
• Workshop: Our first NTU AI Safety Workshop attracted 69 applicants across two cohorts, with 48 final participants after screening.
• Speaker events: We have hosted three speakers, including a researcher at METR, a researcher at Anthropic, and an incoming DPhil student in AI safety at Oxford.
• Technical AI Safety Workshop (12–13 September): We launched a technical AI Safety workshop covering jailbreaks, emergent misalignment, training safer AI, and a hands-on jailbreak hackathon.

What are the most likely causes and outcomes if this project fails?

Cause: The main constraint is organizer capacity. NTU AIS can continue operating without additional funding, but doing so would require organizers to bear a greater financial and time burden themselves. Last semester, only four members were responsible for preparing and running the weekly reading group, resulting in a heavy workload that pushed our capacity to its limit. We therefore plan to expand the organizer team and recruit facilitators to distribute the workload and make the program more sustainable. While student organizers can contribute significant time and energy alongside their studies, they also need to balance coursework and research. A modest stipend would reduce the need to take on additional work to cover living expenses, allowing them to dedicate more time to building NTU AIS and sustaining its activities.

Outcome: Without additional support, the reading group may need to revert to two tracks or move to a biweekly format. Organizers and facilitators may also have less time for their own learning, research, and other commitments. Moreover, limited time, workload, and reduced momentum may prevent us from applying for fellowships or attending EA and AI Safety events. This would limit our ability to bring new knowledge and experience back to the community, as well as provide career advising, which is one of the most important components of our program. Over time, these constraints could also lead to organizer burnout and slow the growth of the AI Safety community.

How much money have you raised in the last 12 months, and from where?

In the last 12 months, we have received the following grants:

• [Finished before August 2026] Kairos Pathfinder Support Grant, Spring 2026 — $1,980: Covered speaker fees and compute resources.

• [Finished before June 2026] Taiwan AI Center of Excellence (Taiwan AICoE) — $5,000: Local government sponsorship for the previous semester. Due to strict per-meal spending caps and other funding restrictions, only around $900 of the $5,000 budget was ultimately usable for our activities. The funding has now ended and did not support stipends.

• [Finished before September 2026] BlueDot Rapid Grant, September 2026 workshop — $1,500: Covers workshop stipends and partial compute resources. We just finished the workshop on Sep 12th and 13th.

• [For Fall 2026 semester] Kairos Pathfinder, Fall 2026 — $ 12,000: Covers most expenses aside from stipends.