Help ensure advanced AI goes well.
AI may become the most transformative technology of this century. Ensuring it is developed safely is an open research problem — the field is young, the stakes are high, and technically strong students can contribute early.
Why AI safety matters
-
Progress is fast
Frontier AI systems now match or surpass human experts across a growing range of tasks, and the leading labs are explicitly working towards more general and more autonomous systems.
-
Alignment is an unsolved problem
We do not yet know how to reliably make advanced AI systems pursue the goals we intend, or how to verify that they do. Many leading researchers — including Turing Award and Nobel laureates — consider catastrophic risks from advanced AI a serious possibility worth addressing now.
-
The field is neglected
Far more people work on making AI more capable than on making it safe. For someone early in their career, that gap represents unusual leverage.
For the full argument, see 80,000 Hours' problem profile: Preventing an AI-related catastrophe.
Opportunities as a KTH student
The field needs people who can do rigorous mathematics, build real systems, and reason carefully about how they fail.
Technical research
Interpretability, evaluations and alignment theory. Backgrounds in computer science, mathematics and physics translate directly — including thesis projects.
Engineering
Safety teams need strong engineers for evaluation infrastructure, red-teaming tools and research platforms — often more urgently than theorists.
Governance & policy
Standards, audits and international coordination. Policy work needs people who understand the systems being regulated.
Introductory course in AI safety — Period 2
This autumn we plan to run an introductory course in AI safety during Period 2: weekly readings and small-group discussions covering the core arguments, the current research landscape and career paths into the field. No machine-learning experience is required.
Registration is non-binding. We will contact you when applications open.
Resources & communities
A short, curated list — each is a good next step depending on where you are.
- AISafety.com → An overview of the entire field: courses, communities, funding, jobs and events.
- Stockholm AI Safety → Local meetup with talks and discussions on beneficial AI, every other Wednesday evening.
- BlueDot Impact → Free, cohort-based online courses — including AI Safety Fundamentals — used by thousands as their entry point into the field.
- 80,000 Hours: AI risk → The most thorough written case for AI risk as a pressing problem, with career reviews and a job board.
- Alignment Forum → Where researchers publish and debate current alignment work. Technical, and close to the research frontier.
New to effective altruism?
AI safety is best understood in context: how does it compare with other ways of doing good? Our intro course covers exactly that question — applications are open now.