Can AI sovereignty strategies prevent overconfidence in jagged AI capabilities?

What is 'jagged AI' and why does overconfidence matter?

Jagged AI refers to the uneven, unpredictable performance of AI systems—they excel at some tasks and fail at others, often in ways that aren't obvious until you test them. Overconfidence in these capabilities means assuming an AI can handle a broader range of tasks than it actually can, which can lead to costly mistakes in critical areas like defense, healthcare, or economic planning. The EU's analysis of its own AI sovereignty reveals that this overconfidence is a real risk: frontier AI models come almost exclusively from the US or China, and the EU holds only 15% of global hyperscale data center capacity [1]. If European policymakers assumed they could rely on these foreign models without understanding their limitations, they'd be overconfident in a capability they don't fully control or understand.

The UK policy brief reinforces this by pointing out that gaps between different policy areas—research, industrial strategy, and procurement—create vulnerabilities and strategic dependencies [2]. When a nation's AI capabilities are 'jagged' (strong in some areas, weak in others), overconfidence can arise from focusing only on the strengths while ignoring the gaps. Sovereignty strategies that map these gaps force a more honest assessment.

How do sovereignty strategies expose blind spots and prevent overconfidence?

AI sovereignty strategies work by systematically breaking down the entire AI value chain into layers and components, then identifying where a nation is dependent on others. The EU framework does exactly this: it decomposes the frontier AI stack into 5 layers, 26 components, and 29 sub-components, then maps these against 5 sovereignty pillars (economic competitiveness, resilience, security and defense, European values, and foreign relations) [1]. This structured approach reveals critical gaps and inter-pillar trade-offs that a narrow economic lens would miss. For example, the EU's AI Gigafactory Initiative looks good for competitiveness but creates conflicts when viewed through the security or values pillars [1]. By making these trade-offs explicit, the strategy prevents policymakers from assuming that a single investment solves all problems—a classic source of overconfidence.

The UK report similarly argues that an integrated approach connecting research, industrial strategy, and procurement is needed to build 'enduring capacity rather than temporary access to technology' [2]. Temporary access (like renting cloud compute from a foreign provider) can create a false sense of capability—you can run the model today, but you don't control its future development or limitations. Sovereignty strategies that prioritize building domestic capacity force a more realistic view of what you actually own and understand.

What are the limitations? Can sovereignty strategies themselves create overconfidence?

Yes, sovereignty strategies can backfire if they are too narrow. The EU framework explicitly warns that focusing only on economic competitiveness—without considering resilience, security, values, or foreign relations—can obscure conflicts and create new vulnerabilities [1]. For instance, a nation might pour resources into building its own foundation model (a 'moonshot' approach) while neglecting data sovereignty or ethical governance, leading to overconfidence in a homegrown system that is actually fragile or misaligned with public values. The UK report notes that there are many possible sovereignty strategies, from developing models from scratch to simply licensing foreign technology [4], and each carries different risks of overconfidence. Licensing gives you access but not understanding; building from scratch gives you control but may be unsustainable.

Data sovereignty—ensuring data is used under agreed terms—is another piece of the puzzle. A 2022 study on data-sovereign AI pipelines found that such components can reduce barriers and increase success in collaborative data science [3], but the study also identified three barriers to implementation [3]. If a sovereignty strategy focuses too heavily on data control without addressing compute or talent gaps, it could create overconfidence in data-rich but compute-poor capabilities. Finally, the ethical dimension matters: one paper argues that states and big tech have fundamentally different goals (peace and security vs. profit) [5], and sovereignty strategies that ignore this tension may overestimate their ability to govern AI in the public interest.

About These Sources

This answer is built on 5 studies (3 peer-reviewed, 2 preprints) — published from 2021 to 2026, 2 from 2024 or later — selected as the most relevant from 5 studies that passed quality screening, drawn from 33 papers retrieved from a database of over 500 million.

Sources used in this answer

Related research questions