As the generative artificial intelligence boom enters its sobering phase, the gap between market hype and geopolitical reality has never been wider. Major firms continue to chase inflated valuations through questionable tokenomics and ever-larger models, even as real returns diminish and logistical strains mount. Policymakers in the Indo-Pacific and across Western Europe and North America (WENA) are waking up to a costly illusion: mistaking fluent, confident outputs for reliable intelligence, in what can be called “perception dysmorphia.”
This misperception is a distorting strategy. Governments remain fixated on the visible hardware contest—chips, energy, rare inputs, and supply-chain chokepoints, while paying far less attention to the deeper, less tangible layers of AI ecosystems: model behavior, data governance, benchmark integrity, and normative influence. The result is a strategic imbalance. States are over-securitizing inputs and under-governing outputs, leaving the most consequential domains of AI power largely unregulated and open to capture
Recent shocks illustrate the limits of a purely resource-centric lens. Ongoing geopolitical tensions involving Iran, and disruptions near the Strait of Hormuz, have constrained global helium flows, affecting the production of high-bandwidth memory (HBM). While peripheral materials can acquire strategic weight within tightly coupled technological ecosystems, they follow geoeconomic rather than AI-centric geopolitical logic. Nevertheless, an apparent geopolitical risk around AI lies in the interdependence of minor inputs across complex supply chains. Firms hollow out domestic oversight by offshoring infrastructure to scale and re-import controversial systems, shifting the strategic competition toward normative evasion.
In the TikTok case in the United States, executive action forcing a partial divestment turned into a broader geopolitical contest over data flows and sovereign control. Similar dynamics are increasingly visible across Southeast Asia in countries like Vietnam and Indonesia, where regulatory ambition often outpaces enforcement capacity. Elsewhere, in India, debates are shifting towards more enforceable, contract-based frameworks.
The Weaponization of Regulatory Capture
Regulation, in principle, acts as a defensive tool to shield local markets. As noted by others, it is fundamentally a public interest function. Yet the dialogue between developers and policymakers remains fractured. Across both WENA and the Indo-Pacific, traditional industry forums and Track 1.5/2 mechanisms – the semi-official and unofficial diplomatic channels typically relied upon to bring government decision-makers and non-state experts together for collaborative, off-the-record problem-solving – rarely reflect the concerns of AI start-ups or technical communities. Governments continue to treat AI adoption as a problem of scale — access, compute, deployment – rather than one of tech & ecosystem resilience. This misdiagnosis is increasingly visible across talent markets and academic systems alike.
Notably, at the International Conference on Machine Learning 2026 (ICML 2026), nearly 795 peer reviews were withdrawn after reviewers were found to have used LLMs despite explicit restrictions. While numerically limited, these cases expose early cracks in the integrity of AI research systems. Similarly, the 2025 “talent wars”, in which firms such as Meta and xAI offered compensation at “professional sports” levels, quickly collapsed into restructuring cycles and high turnover. At the same time, “benchmark capture” is reshaping the regulatory landscape. Nearly 46% of AI safety benchmarks were produced between 2023 and 2024, often by the same entities developing the models. This shows that AI systems are optimized to pass tests rather than demonstrate genuine safety. Large firms shape the standards they already meet, effectively capturing the regulatory narrative, while smaller actors face disproportionate compliance burdens calibrated for industry giants, hinting that safety risks can become a performative metric.
As the hype cycle cools, statecraft is beginning to move beyond tool-centric technological solutionism. In the Indo-Pacific, the recalibration and decline of non-profit and foreign tech actors in development initiatives reflects this shift. For instance, the 2026 Indo-Pacific Forecast from the Center for Strategic and International Studies shows that regional dynamics prioritize counter-coercion and supply chain resilience over technological optimism.
Nevertheless, without sustained investment in AI regulatory literacy and technical training, these frameworks struggle to translate into effective policy. Delegating governance to algorithmic systems risks compressing complex political judgement into probabilistic outputs, ultimately weakening state capacity and strategic autonomy.
Against this backdrop, three trends are likely to emerge over the next 6–12 months:
First, economic diplomacy negotiations among minilateral and bilateral groups will largely decouple from AI and cross-border data governance issues, or treat them as peripheral rather than central concerns. For instance, the European Union’s (EU) AI Act’s high-risk AI obligations are not expected to apply until late 2027.
Second, dedicated technology diplomacy tracks in AI are inevitable to be convened but will bifurcate along industry lines rather than converging. One track will be driven by AI and IT services, prioritizing cross-border data flows and interoperability. The other will center on AI in manufacturing, focusing on supply chain resilience and industrial policy. These competing logics will shape distinct diplomatic alignments, including within Southeast Asia, where economies especially Vietnam (for instance) must navigate between service-led digital integration and manufacturing-led strategic positioning.
Third, AI regulation will evolve in two sequential phases: first, consolidating long-term research agendas for AI safety within information security and legal communities; second, developing contract-based intellectual property protections and case-by-case governance measures as extensions of traditional data governance. Whether these frameworks remain reactive or become anticipatory will depend on how technical standards emerge.
Conclusion
To build a durable diplomatic architecture, policymakers must move beyond abstract risk categories and focus on observable impacts such as documented system failures, structural shifts in market organization, and the cascading economic effects of automation. Filtering and foolproofing the future of AI research trends is now a political necessity for effective AI awareness among governments. Statecraft must therefore decide based on real AI trends amidst the chaos of geopolitics, not just resource geoeconomics.
Abhivardhan is an AI Strategy & Governance Specialist. He is the President of the Indian Society of Artificial Intelligence and Law, and leads a research firm, Indic Pacific. His interests and expertise span Indo-Pacific studies, AI governance, digital competition and AI & intellectual property strategies.
Genevieve-Donnellon May is a Vasey Fellow at the Pacific Forum in the United States and a non-resident fellow at the Yokosuka Council on Asia-Pacific Studies (YCAPS) in Japan. Genevieve is also an associated fellow at the Institute for Security Development Policy in Sweden.
