AI governance today is fragmented across multiple multilateral platforms. Critically analyse the challenges this poses for coherent international regulation, with reference to BRICS' recent initiatives.

Artificial intelligence is today governed by a patchwork of overlapping forums — the EU AI Act, the G7 Hiroshima Process, UNESCO's AI ethics framework, and now BRICS. At its 18th Summit in New Delhi (12–13 September 2026), BRICS adopted the New Delhi Declaration 2026, committing members to a separate statement on "Global Governance of Artificial Intelligence" [1]. Such plurality broadens participation, but fragments regulatory coherence.

Challenges posed by fragmentation

• Regulatory divergence: Competing standards on risk classification and liability raise compliance costs for cross-border AI firms and invite forum-shopping.
• Soft-law deficit: The BRICS commitment is declaratory — "committed to implementing" — not treaty-based, so outcomes depend wholly on national follow-through [1].
• Geopolitical bloc formation: A BRICS track risks being read as a counterweight to Western-led frameworks, hardening a technological bipolarity that impedes consensus on safety norms.
• Diluted agenda focus: AI sits bundled with cybersecurity, DPI, quantum and telecom in the Declaration, weakening depth on AI-specific safety [1].
• Duplication and confusion: India hosted two distinct "New Delhi" AI declarations in 2026 — the AI Impact Summit's Declaration (February) and the BRICS Declaration (September) — illustrating overlapping tracks [2].

The convergent potential

• Equity gains: BRICS' call for wider access to AI resources addresses the compute divide facing developing nations [1].
• Bridging role: The New Delhi Declaration on AI Impact, endorsed by nearly 90 countries and organisations spanning the Global North and South under "Sarvajan Hitaya, Sarvajan Sukhaya", shows fragmented tracks can still converge [2][3].
• Shared substantive concerns: Trustworthy AI and energy efficiency of AI systems recur across frameworks, providing common regulatory ground [1].

Fragmentation therefore reflects not merely rivalry but differing developmental priorities. The way forward lies in interoperability rather than uniformity — mutual recognition of safety standards, a UN-anchored coordinating mechanism, and India using its convening capital to align Global South equity concerns with Global North safety norms. An inclusive, plural yet interoperable architecture best serves AI's promise as a global public good.