This application is a spinoff from a previous grant application we submitted on grantmaking.ai

Project summary

AI Safety Hong Kong (AISHK) is Hong Kong's first (and so far, only) dedicated AI safety fieldbuilding organisation.

Hong Kong is substantially picking up its AI activity, but so far has not been commensurately investing in AI safety. Most of the Government's attention is on AI capability development and diffusion. Any little discussion of risks tends to center around issues like bias, privacy, and transparency. Catastrophic and existential risk from frontier systems is barely present in the public awareness.

China is one of the two frontier AI powers in the world, and yet the global AI safety movement has almost no connection to Chinese researchers, labs, and governance institutions. Some organizations like Concordia AI and SAIF do valuable bridging work, but there is very little direct fieldbuilding from inside the Chinese industry itself. There can be no AI safety if the field remains concentrated around US-led, Western AI industries.

Cities like Beijing, Shenzhen, Hangzhou, and Shanghai may host more of China's leading labs and technical talent, but Hong Kong is where we think fieldbuilding has the most leverage on the Chinese AI industry as a whole. It is being built up as the financial heart of the Chinese AI industry, where frontier AI labs incorporate, raise capital, and launch globally, all while the city retains its own legal and regulatory system, separate from the mainland's. The Hong Kong government is actively working to make Hong Kong the bridge between mainland China and global markets, and to establish it as the "superconnector" of the Chinese Greater Bay Area, a role written into its current Five Year Plan. We want AI safety to be an integral part of that role. In fact, a growing number of our event attendees already cross the border from Shenzhen to join us.

Our long-term goal is a permanent AI safety hub in Hong Kong, integrated into both the Western and Chinese AI industries, where people can build careers, run research programmes, and launch organizations aimed at tackling frontier AI risks.

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

Where we want to be in 12 months:

• Reading groups running at 40+ capacity instead of turning half our registrants away, plus a second monthly format to absorb the overflow.

• Two intro fellowship cohorts delivered, 10–15 participants each. Five weeks, adapted from the BlueDot curriculum with additional readings for a Chinese and regional perspective. First cohort launches late 2026.

• The ML4Good and TARA Hong Kong cohorts successfully delivered, and at least one further training or fellowship programme brought to Hong Kong. We are currently in discussions with several well-known capacity-building and fellowship programs.

• Student AI safety societies running at two or three more Hong Kong universities beyond CUHK, feeding a talent pipeline into the field.

• Three to four research and policy outputs published, including work on AI in Hong Kong's financial sector and on the gaps in the government's AI strategy.

• Hong Kong's first dedicated AI safety conference held, mid-2027, at target funding.

• Monthly webinars continuing, with speakers already confirmed from Anthropic, the Future of Life Institute, the Oxford Martin AI Governance Initiative, and Hong Kong universities.

How will this funding be used?

Full budget here: https://docs.google.com/spreadsheets/d/1BdiAR6BJ4I0NYTilkV-er2S3FH-b2pQ16GhOxcmH3kc/edit?usp=sharing

We have structured the request in three tiers.

$20,000 (minimum). Runs our current programme for another year: digital infrastructure, monthly in-person events and study circles, basic incorporation, a coworking subspace, advocacy outreach, and marketing.

$60,000 (scale-up). Lets us grow to meet demand. Larger venues and catering for our reading groups and other socials, a dedicated 12-month team room that doubles as our event space and registered address, full incorporation as a company limited by guarantee with a Section 88 charity application, a funded research and policy programme, Chinese-language localisation of our materials, and expanded marketing.

$85,000 (target). Adds Hong Kong's first dedicated AI safety conference in mid-2027, scoped as a first edition (~150 attendees, 1-2 days, 4–6 international speakers), and travel support for the team to represent AISHK at international AI safety conferences.

We will not be using funding for stipends for the core team for our first year or so of operations, and will continue to run AISHK on a volunteer basis for the duration of this grant.

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

AISHK's unfunded six-month track record:

• Monthly webinars since April. Live attendance usually averages 50+ participants from all over the world. Upto 200+ registrations per event.  Our webinar calendar can be found here.

• Monthly in-person reading groups at 20–25 attendees, averaging 40+ registrations for 25–30 places. Our reading group calendar can be found here.

• 400+ Substack subscribers at a 55% open rate. Our subscriber base is growing organically off of event attendance, usually increasing by 70-100 new subscribers per month. Our Substack can be found here.

• Collaborating with both ML4Good and TARA as their Hong Kong partner for their first cohorts in the region.

• We're helping launch an AI safety group at CUHK, with interest from other Hong Kong universities.

• We recently submitted a representation to the Hong Kong Constitutional and Mainland Affairs Bureau in response to their Five Year Plan consultation document.

Our team:

Kenneth Chan is Managing Director of Effective Altruism Hong Kong. He founded GiveWise, Asia's first tax-efficient effective giving platform, and has spent a decade building high-impact communities. He gathered together the core team to start AISHK, and drives day-to-day operations.

Aya Lowe is a technology policy and digital safety advisor with fifteen years of experience across APAC and the Middle East, spanning government, media, and the technology sector. She has worked with Meta on misinformation and crisis response, and with Australia's eSafety Commissioner. She leads communications.

Chris Barnes is a Risk Technology Product Lead in Hong Kong’s banking sector and a dual-track alumnus of Bluedot Impact’s Governance and Technical Safety programs. With AI Safety HK, he applies his deep expertise in systemic risk mitigation and product leadership to the complex challenge of AI safety.

Abeer Sharma holds a PhD in Law from HKU and has practised before the Supreme Court of India and international tribunals. He has advised sovereign states, regulators, and technology enterprises on governance design. He leads research, policy, and communications.

Tzu Kit Chan is our advisor. He is Chief of Staff at Atlas Computing, ran Stanford AI Alignment, cofounded Caltech AI Alignment, did operations for MATS, and has advised more than forty-five university and national AI safety groups worldwide.

We are also supported by Alice Wong (Assistant Lecturer, HKU), Hayden So (Head of Strategic Initiatives, FarmWorks, previously Hong Kong government), and Baber Rehman (AI Compiler Engineer, Huawei HK).

All of our team members have had extensive prior engagement with AI safety (e.g., via Bluedot courses, AI safety fellowships, research collaborations, etc).

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

• Key-person dependency: We run on a handful of volunteers around full-time jobs. Momentum is dependent on enough volunteers freeing up spare time from their day jobs.

• Audience without conversion: It's possible that we draw large audiences (attracted by the "AI" label) but see few of them go to actually work in AI safety specifically. The intro fellowship and student groups are meant to move people past that first-exposure stage. So far we've been having a decent conversion rate for people getting interested in AI safety specifically.

• Talent leakage: People we upskill might take safety roles abroad, leaving Hong Kong without a resident ecosystem of its own.

• Transferability barriers: We might stay a small English-speaking bubble that never reaches the Greater Bay Area or the Chinese industry the way we're aiming to. We are, in fact, already in discussions with other China-focused AI safety groups that are translating popular educational content into Chinese.
  
  The risk of failure is quite modest. Our financial costs are low, and we've already built significant enough traction to inspire local groups and high-impact professionals to start exploring AI safety.

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

We have not raised any money so far, and have been covering our operations with personal funds.

A note on compliance

AISHK's work is limited to public education, research, and convening on AI safety and governance. We operate with full transparency: our events, materials, and research outputs are public, and our policy engagement runs through official government consultation channels rather than lobbying. None of our activities contravene any applicable national security laws.