Trace the evolution of global AI governance summits from Bletchley Park (2023) to New Delhi (2026). How has the framing shifted from existential risk to real-world harms?

The four-summit arc — Bletchley (2023), Seoul (2024), Paris (2025), New Delhi (2026) — records a steady migration of global AI governance from speculative, frontier-model catastrophe towards present-day questions of access, equity and developmental impact.

Bletchley 2023: the existential-risk frame

• The Bletchley Declaration, signed by 28 countries and the EU, centred on "frontier AI" and the potential for "serious, even catastrophic, harm", especially misuse and loss of control [1].
• Concerns were framed around cybersecurity, biotechnology and disinformation — hazards of capability, not of distribution [1].

Seoul 2024: institutionalising safety

• The Seoul agenda widened the triad to safety, innovation and inclusivity, adding transparency and risk-management commitments from frontier developers [2].
• A network of national AI Safety Institutes was seeded, converting declaratory concern into scientific machinery [2].

Paris 2025: from safety to action

• Renamed the AI Action Summit and co-chaired by France and India, it produced a statement on inclusive and sustainable AI for people and the planet, signed by 62 countries — prioritising digital divides, openness and sustainability [3].
• Notably, the US and UK declined to sign, exposing the safety-versus-acceleration cleavage [3].

New Delhi 2026: real-world harms and the Global South

• The first such summit hosted in the Global South, organised by MeitY around three sutras — People, Planet, Progress [4].
• The New Delhi Declaration on AI Impact, endorsed by 92 countries and organisations, rests on human capital, social empowerment, trustworthiness, energy efficiency, AI for science, democratised compute and economic good — a harms-and-benefits agenda, not a doom agenda [5].
• UN Secretary-General Guterres anchored the equity turn by proposing a $3 billion global fund for skills, data and affordable compute in developing countries [6].

The shift reflects both maturing evidence — everyday harms like bias, job displacement and data asymmetry proved more immediate than machine takeover — and the entry of developing-country voices. Sustaining it requires that New Delhi's commitments translate into financed compute access, consistent with the Global Digital Compact and SDG-9.