The map: institutions, arguments and money
The landscape of AI safety is complex and fragmented. It is shaped by overlapping institutions, competing research agendas, and shifting financial priorities. Understanding how these elements interact is essential to navigating the current state of the field.
This lesson maps the three core layers of that landscape and explains why they depend on one another.
The three layers of the AI safety ecosystem
AI safety in 2026 is best understood as a system of three interconnected layers. The first layer consists of institutions responsible for testing and regulating AI systems. These include safety institutes, regulators, and standards bodies.
The second layer is composed of research schools that debate the nature of the problem itself. These groups shape the intellectual direction of the field. The third layer is made up of funders who determine which arguments receive financial and institutional support.
This tripartite structure is not static. The relationships between these layers shift as new challenges emerge. For example, the AI Safety Institute became the AI Security Institute in 2025.
This institutional evolution reflects the growing emphasis on security over safety in policy discussions. Similarly, ARIA’s programme was restructured in 2026. These changes highlight how the field’s institutions adapt to new realities.
The consensus artefact: the International AI Safety Report
A key coordination mechanism in the AI safety field is the International AI Safety Report 2026. Published on 3 February 2026, this document brings together over 100 experts from more than 30 countries and organisations. These include the European Union (EU), the Organisation for Economic Co-operation and Development (OECD), and the United Nations (UN).
The report was chaired by Yoshua Bengio, a Turing Award winner. It is the closest thing to an IPCC report in the AI safety domain.
This report is significant not only for its scope but also for its role as a benchmark. It synthesises findings from across the field, offering a shared reference point. It is a consensus artefact that helps align efforts despite the lack of formal coordination.
The report’s publication marks a turning point in how the field defines and communicates its priorities.
UK leadership in AI safety infrastructure
The United Kingdom stands out as a central hub in the AI safety ecosystem. It hosts the AI Security Institute, which is the largest government AI evaluation capability in the world. The UK also hosts ARIA’s Safeguarded AI programme, a formal-methods initiative that has drawn significant attention.
In addition, the country is home to the Turing Institute and the AI Standards Hub. These institutions form a dense network of expertise and policy influence.
The UK’s regulatory approach further distinguishes it. Rather than a single AI Act, the country has adopted a sectoral regulatory architecture. This means that different sectors are regulated by different bodies.
This approach allows for more nuanced oversight but also creates coordination challenges. It reflects the UK’s broader strategy of blending innovation with risk management.
Dynamic change in structure and identity
The AI safety landscape is not only divided into layers but also in constant motion. Institutional names and functions change rapidly. For example, the AI Safety Institute became the AI Security Institute in 2025.
The ARIA programme was restructured in 2026. The original Frontier AI Taskforce has already institutionalised twice.
These shifts reflect the field’s high pace of development. They also show how quickly new priorities can reshape the ecosystem. The frequent realignment of roles and responsibilities indicates a field in flux.
It is a landscape where even the most recent developments can quickly become outdated.
Building a navigable map of the field
To be useful, any map of the AI safety landscape must be more than a list of names. This course provides a navigable framework by naming specific repositories, reports, and funding calls. It offers links to resources that readers can actually access.
This approach ensures that the lesson is not only descriptive but also actionable.
The course is structured as part of a sequence of five. It is the third in the sequence, building on earlier lessons on plain-English safety and safe adoption. It leads into more advanced material on hands-on evaluation and provable-safety research.
This ladder position makes it a critical bridge between foundational understanding and technical application.
What to take away
The AI safety field is structured around three interacting layers: institutions, researchers, and funders. The International AI Safety Report 2026 is a key consensus artefact in this landscape. The UK plays a uniquely central role, hosting several key institutions. Finally, the field changes rapidly, and its structure is in constant flux. Understanding these dynamics is essential for anyone seeking to engage meaningfully with the field.
Reference
| Lesson | 1 of 15 |
| Outcome | Describe the three layers of the AI safety landscape and why each needs the others. |
| Consensus baseline | International AI Safety Report 2026 |
| Frontier evidence | AISI research index |