AI Safety Diliman

AI Safety Diliman is a Quezon City-based AI-safety community organization founded in 2025 at the University of the Philippines Diliman (UPD) campus, building the Filipino talent pipeline for AI safety and governance through two program tracks: an AI Governance Accelerator and a Technical Alignment Program. Supported by Kairos and fiscally sponsored by the Berkeley Existential Risk Initiative (BERI), the organization's stated mission is to grow "the community of people working on the most pressing AI safety challenges in the Philippines, in Southeast Asia, and around the world." It is the corpus's first Philippine entry and the sole Philippine group listed in the AISafety.com global communities directory.

Founding and context

AI Safety Diliman launched in 2025 from the University of the Philippines Diliman — the flagship campus of the country's national university system in Quezon City, Metro Manila — and describes its founding rationale in terms of the Philippines' structural position as a consumer of frontier AI technologies: the country "imports much of its most critical digital infrastructure," making local AI-safety capacity especially important. The organization emerged from the EA UPD and EA Philippines community ecosystem, and operates alongside Effective Altruism Philippines as a distinct organization with its own program structure and identity.

Its fiscal infrastructure reflects the global EA-adjacent AI-safety field-building network. Kairos, a US nonprofit running the Pathfinder Fellowship (which funds and mentors student AI-safety organizers worldwide) and the SPAR research fellowship, provides operational support; BERI provides fiscal sponsorship — the same arrangement used by many AI-safety local groups that have launched in non-Western hubs over the 2023–2026 period.

Programs

AI Safety Diliman runs two program tracks targeting different talent segments:

AI Governance Accelerator. A 10-week cohort-format program open to students and early-career professionals, building foundational capacity in AI governance with particular attention to the Philippines' and Southeast Asia's position as consumers of frontier AI systems. The curriculum moves through technical foundations — LLMs, the AI supply chain, scaling laws — to the governance toolbox (alignment research, safety evaluations, compute governance, regulatory frameworks) to governance strategies for consumer states: where countries without frontier-development capacity actually have leverage. The accelerator culminates in small-team research projects. Cohort 1 concluded with a Demo Day on 24 May 2026 at SevenSpace on Maginhawa Street in Quezon City; the three projects presented took locally grounded approaches — a study of how LLMs handle politically sensitive Philippine topics (using Sara Duterte's impeachment as a case study), a BPO AI Worker Dialogue Lab examining how AI tools reshape workflows and workplace surveillance in the Philippines' large business-process-outsourcing sector, and a legal framework initiative titled "From Principles to Black Letter Law" translating abstract AI safety principles into enforceable mechanisms for countries that consume rather than develop frontier AI.

Technical Alignment Program. A semester-long intensive for participants with technical backgrounds, developing skills in mechanistic interpretability, scalable oversight, and technical AI safety research — the capability pipeline for participants targeting research roles in the field.

Intro events. The organization also runs public introductory programming; the first Intro to AI Safety talk ran on 30 January 2026, framing the Philippine perspective as "more important than ever" in the field.

Place in the movement

AI Safety Diliman is the corpus's first Philippine entry and the first Southeast Asian AI-safety community group in the corpus beyond the Singapore AI Safety Hub (SASH). It sits in the insider-community pattern alongside Berlin AI Safety, AI Safety 東京, AI Safety South Africa, and AI Safety India — EA-adjacent groups building the local practitioner pipeline for AI-safety work, supported through the Kairos / BERI infrastructure.

What distinguishes AI Safety Diliman's framing is the explicit consumer-state governance angle: where other groups in this pattern tend to foreground technical alignment capacity-building, AI Safety Diliman explicitly centers the governance perspective of a country that imports its most critical AI infrastructure rather than produces it. The Cohort 1 research projects reflect this framing directly — the BPO worker dialogue lab and the legal framework initiative both address Philippine institutional conditions rather than adapting Western frameworks wholesale. This positions AI Safety Diliman as part of the emerging current within global AI-safety field-building that asks how non-producer countries can exercise meaningful governance leverage — a question increasingly central as the gap between frontier-model developers and the broader world's ability to govern those models widens.