[SAIPH] Incidents to interventions: Frontier AI safety research and lessons for non-frontier resilience-building

​What can an AI safety incident at the frontier teach a country that doesn't build frontier AI?

In the Hugging Face incident, AI agents exploited weaknesses in shared infrastructure and ageing web software. The task of detecting, containing, and piecing together what happened fell largely to security engineers and software engineers, without the involvement of alignment researchers. As AI agents spread through the systems that Filipinos build, secure, and regulate, countries far from the frontier will have to decide how to prepare for incidents like this one.

Safe AI Philippines is hosting a three-hour, in-person hybrid workshop for students and early- to mid-career professionals interested in learning more about frontier AI safety research and applying its lessons to resilience-building in the Philippines.

​Our speaker is David Demitri Africa, a research scientist at Resolution, MATS mentor, and former alignment researcher at the UK AI Security Institute. David will use the incident as a running case to show what technical AI safety and security research at the frontier looks like.

Afterwards, you'll work through the incident with peers from your own field and take on a harder question. Of everything a country like the Philippines could do, what is the single most impactful intervention, and how would you choose it? The goal is for you to leave with concrete next steps into AI safety work!

Agenda

• ​4:00 - Arrival

• ​4:15 - Briefing on the Hugging Face incident for a non-technical audience

• ​4:30 - Talk by David Demitri Africa: Doing Technical AI Safety and Security Research at the Frontier

• ​5:05 - Open forum

• ​5:25 - Break

• ​5:35 - Reflection and cross-field pairs

• ​5:50 - Field-based group discussions

• ​6:30 - Next steps and pathway tables

• ​6:50 - Open networking

• ​7:00 - End of event