Open at the interface, closed at the core: neutrality claims in AI‑enabled geoscience infrastructure focusing on Africa – The case of Deep‑Time Digital Earth and GeoGPT

Paul Hugh Cleverley, Ezzoura Errami, Tania R Marshall, Simon Burnett, Andrew O Lator, Ebuka Ibeke

Geological data has become strategic infrastructure for resource assessment,
investment negotiation, and economic sovereignty. Remotely hosted big data and
Artificial Intelligence (AI) platforms are increasingly promoted for geoscience in the
Global South, particularly Africa. Whether such platforms are genuinely geopolitically
unaligned, and how this can be verified, has direct implications for scientific autonomy
and data sovereignty.

Using Deep‑time Digital Earth (DDE) and GeoGPT as a critical case, this paper examines
whether claims of geopolitical non‑alignment can be independently verified. DDE was
suspended by its founding body, the International Union of Geological Sciences (IUGS),
in early 2025, yet its non‑alignment claims have continued. Drawing on UNESCO Open
Science and AI Ethics frameworks, the paper assesses three criteria: governance
independence, data and software openness, and AI transparency.

The analysis reveals a distinction between user‑interface accessibility and core technical
verifiability. Core DDE/GeoGPT software remains proprietary, and the platform and
models diverge from definitions of ‘openness’ and ‘digital public goods’ established
by UN‑backed frameworks, and the AI Ethics Recommendations from UNESCO. The
documentary record also shows governance and funding concentrated in a single
national jurisdiction, including links between DDE collaboration targets and China’s Belt
and Road Initiative, a 2026 agreement with the mining investment department of the
China‑Africa Development Fund, and advisory structures that include individuals with
documented affiliations to the national resource‑extraction sector. Combined, their
significance lies in the gap they create between the platform’s claims and what can be
independently verified.

AI governance documentation discloses filtering of content deemed politically sensitive; in
one documented instance, a response citing peer‑reviewed research on state censorship
was replaced by a refusal. The criteria and scope of this filtering are not fully disclosed,
leaving it un‑auditable.

The paper concludes that it is currently not possible to verify claims of non‑alignment
using this framework; DDE/GeoGPT is best characterised as what we term a
state‑anchored international platform: globally participatory, but dependent on a single
national jurisdiction. The criteria provide a transferable test for comparable claims and
highlight the importance of calibrated engagement with external platforms, and use of
open‑source and sovereign alternatives where privacy or geological data sovereignty
is at stake.