AI Safety 東京 (AI Safety Tokyo)

AI Safety 東京 is a Shibuya-headquartered, Tokyo-anchored AI-safety community group whose central activity is a monthly seminar series / reading group — a benkyoukai (勉強会) — convening academics, students, and working professionals in Japan around the technical and policy literature on advanced AI risk. Founded in January 2023, the group exists "to get academics, students and professionals in Japan thinking about and working in AI safety", whether through technical alignment research, AI-governance policy work, or related contributions. It is the corpus's first local-group entry in East Asia and the first entry whose central activity is an insider AI-safety reading group rather than public-facing campaign or protest organising.

Founding and structure

AI Safety 東京 was started by Blaine William Rogers, a Research Communications Officer at the Tokyo-based Noeon Research, with the explicit ambition of "building a safety community in Tokyo from zero" and ultimately serving as an AI-safety hub for Asia — finding interested talent in Japan and routing it into international safety research. Rogers funded the first year of the group's activity out of pocket (visa constraints prevented external grants at the time per his own Manifund disclosure), and the year-one report on Manifund records that the group converted "at least 4 mid career academics and professionals into highly engaged advocates", one of whom went on to work with the Japanese alignment-research organisation ALIGN, another to facilitate a Blue Dot Impact AI Safety Fundamentals Governance course while founding a separate AI-safety best-practices initiative.

The current organiser roster on the group's About page lists Rogers alongside Kaitlan Melone (also Noeon Research), with a wider 30+ member listing of organisers, members, and honored guests that includes Prof. André Röhm (University of Tokyo), Project Professor Ken Mogi (CIRL, University of Tokyo), and senior figures from across the international safety research community. The page names partnerships with the AI Alignment Network (a Tokyo-based AI-safety research organisation) and Effective Altruism Japan — the EA Forum's groups directory lists AI Safety Tokyo Benkyoka within the same Japan ecosystem as the broader EA Japan community.

The benkyoukai — monthly reading group and seminar

The group's central activity is its benkyoukai — described on the dedicated page as a monthly meeting "to discuss a safety-related topic; a paper hot off the presses or a recent advance in policy". Sessions typically run 19:00–21:00 on a Wednesday evening, pitched explicitly at "an interested technical layperson with no assumed domain knowledge", with attendees drawn from computer science, mathematics, physics, law, and philosophy backgrounds. The EA Forum's group listing renders the same format more colloquially as "university seminars for the intro-to-safety course that your university doesn't offer yet". Recent meetings (2026 first half) have covered Scalable Oversight (20 May), the International AI Safety Report (15 April), AGI Timeline Trends following the 2025 reasoning-model releases (25 March), AI Lab Commitments (11 March), and AI Control as a "last line of defense" paradigm (18 February). Recent meeting venues include Kojima Law Offices and the HiRAKU GATE coworking space near Todaimae station.

This format — a low-barrier, technical-but-accessible reading group running on a monthly cadence — is the activity that anchors the group's community-building function: it is, by design, the on-ramp through which professionally-busy mid-career academics and engineers in Tokyo can begin tracking the AI-safety literature without leaving their day jobs, and the surface on which the group's Asia-hub ambition is mechanically expressed.

The Technical AI Safety Conference (TAIS)

Alongside the monthly benkyoukai, AI Safety 東京 co-organises an annual research-conference programme — the Technical AI Safety Conference, or TAIS — with Noeon Research as joint organiser. The inaugural TAIS 2024 ran on 5-6 April 2024 at the International Conference Hall in Tokyo's Plaza Heisei (Koto City), with the AI Alignment Network, the AI Industry Foundation, and Reaktor Japan as collaborators; the programme assembled 18 talks across mechanistic interpretability, scalable oversight, agent foundations, and artificial life, with 105 in-person attendees and approximately 400 unique livestream viewers. The successor TAIS 2025 ran on Saturday 12 April 2025, 12:00–17:00 JST at Tokyo Midtown Tower in Akasaka, again co-organised by Noeon Research and AI Safety 東京 and sponsored by Noeon, Ashgro, and ALIGN; the programme featured workshops by Ryota Kanai (Araya / ALIGN) on global-workspace theory as a route into AI consciousness research, and by Adam Gleave (FAR.AI) on securing advanced AI capabilities. The TAIS 2025 site routes visitors forward to a planned TAIS 2026 edition, indicating the group's intent to maintain the conference as an annual fixture.

The TAIS programme is the group's most visible international-facing surface: where the benkyoukai recruits and trains the local Japanese community, TAIS is where the group convenes that community with international technical safety researchers and gives the wider region a venue for in-person Asia-anchored exchange on AI safety.

Place in the movement

AI Safety 東京 is the corpus's first East-Asia local-group entry and its first entry covering an insider AI-safety community group — one whose central activity is reading-group / seminar / conference work among technical and academic insiders, rather than public-facing protest or campaign organising of the kind that the PauseAI national chapters anchor. The group's audience-shape complements rather than competes with the PauseAI cluster the corpus already covers: where PauseAI chapters in Paris, London, Utrecht, and elsewhere mobilise public protest around a frontier-AI pause demand, AI Safety 東京 cultivates the underlying technical community — the academics, students, and mid-career professionals — within which a serious Japanese AI-safety conversation can take root. As of mid-2026 it sits alongside Tokyo-based safety-research organisations including Noeon Research and the AI Alignment Network, and the Japanese government's own J-AISI fact sheet on AI safety in Japan places that civil-society community-building substrate inside a wider domestic ecosystem that now also includes the institutional safety institute on the government side.