Geopolitics & Sovereign Positioning

Russia is using commercially available American AI tools to automate drone targeting, generate cyberattack scripts, and industrialise disinformation production, according to reporting by Defense One. AI companies have raised alarms through official channels, but the administration has not moved to restrict or sanction the specific vectors of access — whether via third-country re-export, API access through intermediaries, or open-weight models that cannot be recalled once released. This is a qualitatively different problem from semiconductor export controls: chips require physical logistics chains that can be monitored; software and model weights do not.

The strategic implication is significant. The current US export control architecture, built primarily around chip restrictions targeting China, is poorly suited to preventing adversaries from weaponising AI at the application layer. Russia is exploiting this gap in a live conflict. The absence of an administration response suggests either a policy vacuum or a deliberate tolerance, neither of which is a stable position. Separately, The Diplomat frames China's military-civil fusion strategy in AI and robotics as a structural threat, arguing the convergence of autonomous systems and AI decision support is transformational in ways current US posture underweights.

The NSA's cybersecurity chief has confirmed that AI is being deployed to allow analysts to process signals intelligence at speeds previously impossible, compressing multi-day analytical workflows into near-real-time outputs, per Defense One. This is an enacted operational deployment, not a pilot programme announcement. The significance is not merely efficiency: the ability to fuse, correlate, and act on vast data troves faster than an adversary can cycle through decision loops is a structural intelligence advantage. Nations without equivalent AI-augmented intelligence infrastructure face an asymmetric disadvantage in crisis situations and ongoing collection operations.

The NSA deployment also raises second-order questions about reciprocal vulnerabilities. As AI becomes central to US intelligence operations, adversary targeting of the underlying models, training data, or inference infrastructure becomes a priority attack surface. The intelligence community is simultaneously gaining capability and acquiring new systemic dependencies.

Anthropic has published a threat intelligence report confirming it blocked what appears to be a deliberate attempt to use its models to assist in biological weapons development, per BBC News. Simultaneously, a RAND commentary argues that AI systems can now meaningfully accelerate the design of engineered viral threats and that the window for pre-emptive defensive architecture — layered biosurveillance, access controls on synthetic biology databases, model-level restrictions — is closing, per RAND. A Foreign Policy piece advocates for reviving US-China track-two science diplomacy as a mechanism for establishing shared AI biosecurity norms, drawing on the precedent of Cold War scientific exchanges that produced the Biological Weapons Convention, per Foreign Policy.

The Anthropic incident is particularly significant as a public disclosure: it demonstrates that frontier labs now have active threat intelligence functions detecting misuse in near-real-time, but it also confirms these attempts are occurring. The policy architecture governing AI and biosecurity — which currently sits across multiple uncoordinated jurisdictions — has not kept pace. There is no binding multilateral instrument requiring labs to share biosecurity threat intelligence with governments, nor a standardised protocol for what constitutes a reportable incident.

An Atlantic Council analysis makes an explicit case that the US lacks a coherent soft power strategy for Africa on technology and AI, while China has systematically cultivated infrastructure dependencies, training programmes, and regulatory influence across the continent, per Atlantic Council. This is a policy recommendation, not an enacted programme, but it reflects a growing consensus in the US foreign policy community that commercial AI expansion by US firms is not a substitute for state-level engagement. Meanwhile, OpenAI and Anthropic are aggressively recruiting senior executives with Asia market experience from Meta, Google, and Microsoft to lead expansion across India and Southeast Asia, per Rest of World. This signals that frontier labs now treat geopolitical market positioning as a core strategic function.

The dynamics across the Global South are not uniform. In Southeast Asia, US firms are competing commercially against Chinese platforms with stronger established relationships in some markets. In Africa, the contest is earlier-stage and more tilted toward infrastructure and standards influence. African governments retain meaningful leverage as swing states — their adoption of US or Chinese AI platforms, regulatory models, and data governance frameworks will shape which bloc's norms become the de facto global standard for a population that will represent a significant share of global internet users by 2030.

Reporting by Rest of World reveals that China's AI development ecosystem is absorbing large numbers of credentialled but underemployed professionals — lawyers, architects, engineers — as gig workers generating high-quality domain-specific training data. The dynamic is driven by a stagnant economy and state directives that have restructured labour markets, but the strategic output is a pipeline of expert-annotated data that feeds into Chinese AI models. This is not an incidental market phenomenon; it represents a structurally different approach to the data quality problem compared to US labs, which rely more heavily on synthetic data generation and contracted annotation at scale.

The geopolitical implication is that China's economic difficulties are, paradoxically, accelerating one dimension of its AI capability build. The state does not need to centrally coordinate this labour pipeline; economic pressure does the work. Western analysts focused on chip restrictions and compute ceilings should not underweight the data advantage accumulating through this mechanism, particularly for professional and domain-specific AI applications — legal AI, architectural design, engineering simulation — where annotator expertise directly determines model capability.