School four: ethics, FATE and sociotechnical safety

This lesson examines the near-term harms school of AI safety, which argues that current risks from artificial intelligence, such as bias, data exploitation, and power concentration, are more urgent and visible than speculative future risks. It explores the institutional and policy developments that reflect this perspective, including the influence of key research organisations and the growing public and regulatory support for governance-focused approaches.

Core Arguments of the Near-Term Harms School

The near-term harms school contends that the harms caused by current AI systems are already demonstrable and severe. These include algorithmic bias, unfair treatment of marginalised groups, and the exploitation of data workers. The school also criticises the framing of AI safety as a concern primarily for future superintelligence, arguing that such framing diverts resources and attention from pressing present-day issues.

It asserts that governance and regulation, not just technical fixes, are required to address these problems.

This school’s critique is rooted in the belief that the focus on long-term risks can obscure more immediate dangers. It argues that the current systems for evaluating AI safety are insufficiently independent, as much of the benchmarking and alignment work is conducted by the same organisations that benefit from the technology.

This creates an accountability gap that undermines the credibility of current safety measures.

Institutional Carriers of the Near-Term Harms Perspective

Organisations such as the Deep AI Research (DAIR), founded by Timnit Gebru in 2021, play a central role in advancing the study of AI harms outside of corporate influence. The AI Now institute focuses on the concentration of power in the AI sector, while the Ada Lovelace Institute in the UK has made the gap between AI ethics principles and their real-world enforcement a central part of its 2025-2028 strategy.

These institutions are not only producing research but also shaping discourse and policy. The Ada Lovelace Institute, for example, highlights the disconnect between ethical guidelines and actual regulatory frameworks. Their 2026 policy work criticises the UK’s approach to AI governance for relying too heavily on corporate self-regulation and calls for binding legal obligations.

Public and Political Momentum

Public sentiment aligns closely with the near-term harms school. A March 2025 UK survey found that 72 percent of respondents said that regulation would increase their comfort with AI. This sentiment reflects growing public concern over the deployment of AI systems without sufficient oversight. It also signals a shift towards a demand for enforceable standards rather than voluntary measures.

This public support has translated into real policy developments. The Information Commissioner’s Office (ICO) introduced a statutory code duty in 2026. In addition, Ofcom made its first enforcement actions against generative AI providers.

These developments reflect the influence of the near-term harms school in shaping regulatory priorities.

Policy Wins and Institutional Influence

The near-term harms school has seen tangible policy results in 2026. The ICO’s statutory code duty represents a significant step in embedding accountability into AI governance. Similarly, Ofcom’s enforcement actions mark the first time that generative AI systems have faced direct regulatory scrutiny.

These actions are part of a broader legislative framework being debated in the UK, which reflects the school’s influence on the national AI bill.

The influence of this school is also seen in its critique of the technical safety agenda. It argues that benchmarks and alignment work set by AI labs are inherently limited, as they are not externally validated. This critique has gained traction, especially as the limitations of current safety measures become more apparent.

The school’s influence is further evidenced by the inclusion of its concerns in the UK’s AI bill consultations.

Integrating Near-Term and Long-Term Approaches

The near-term harms school does not reject long-term AI safety concerns. Instead, it argues that both areas are important and should be funded and addressed in parallel. The course teaches that conflating these two domains as competing priorities is a false dichotomy.

A comprehensive safety strategy must account for both types of risks.

This integrative approach is reflected in the course’s pedagogy. It does not privilege one school’s perspective over another. Rather, it encourages students to understand how different schools frame the same issues in different ways.

This is essential for developing a complete understanding of AI safety policy and practice.

What to take away

The near-term harms school highlights that current AI risks are both visible and urgent, and that governance must evolve to meet them. It has seen real policy success in 2026, including ICO and Ofcom actions. The school’s influence is also growing in legislative debates, particularly in the UK. The course teaches that addressing both near-term harms and long-term risks is essential for a robust safety framework.

Reference

Sources and further reading

• Ada Lovelace Institute
• Ada Lovelace: Mind the Gap (2025 reflections)