The dispute between the US defence establishment and Anthropic may look like a contract disagreement. It is not. It is an early sign of something more consequential: a struggle over who governs artificial intelligence when it becomes embedded in national power.
Public reporting suggests Anthropic resisted certain Pentagon terms around the military deployment of its models and the scope of acceptable use. The administration responded by directing federal agencies to halt the use of the company’s technology and by designating it a supply chain risk. Sam Altman, CEO of Anthropic’s rival, OpenAI, has indicated that OpenAI maintains similar boundaries in defence contexts. He said, “we have long believed that AI should not be used for mass surveillance or autonomous lethal weapons, and that humans must maintain meaningful control over high-stakes automated decisions. These are our core red lines”. This signals that the tension is beyond partisan politics or procurement procedure, but is more about authority.
|
|
For decades, defence contractors supplied hardware and software to governments that retained ultimate discretion over deployment. The state determined how aircraft, satellites or cybersecurity tools would be used, constrained by law and oversight. Contractors influenced capability, but they did not claim the right to define moral boundaries for sovereign action. Today, artificial intelligence has changed that arrangement.
Frontier AI systems are not inert equipment; they are designed with internal safeguards, usage restrictions and enforceable terms. When companies assert red lines—particularly around military use—they are not merely selling products. They are shaping the operational envelope within which governments can act.
|
|
|
|---|
That shift is historically unusual, and it becomes clearer when placed in context.
The modern internet itself was born from national security imperatives. In the late 1960s, the US Department of Defence funded ARPANET through the Advanced Research Projects Agency (ARPA), later known as DARPA. The goal was to build a distributed communications network that could survive disruption. The architecture that eventually evolved into today’s internet was seeded in military research and public funding.
Over time, however, control diffused. Universities, private firms and civil society actors shaped its evolution. What began as a defence project became global infrastructure, governed through a mixture of technical standards bodies, private platforms and state regulation. The state did not disappear from the internet’s governance, but it ceased to be its sole architect.
AI is following a different trajectory. While governments fund research and shape export controls, the most advanced models are concentrated within a small number of private firms. Unlike ARPANET’s early open architecture, frontier AI is vertically integrated: compute clusters, proprietary weights, internal safety teams and centralised update cycles.
In other words, the infrastructure of cognition is not primarily public; it is corporate.
That concentration creates tension when AI becomes central to national defence. Governments depend on advanced models for intelligence analysis, logistics optimisation, cyber operations and strategic planning. Companies depend on governments for procurement revenue, regulatory stability and geopolitical backing.
The friction between Anthropic and the Pentagon reflects what happens when those dependencies collide.
This is not uniquely American.
In Israel, Project Nimbus—a cloud contract involving Google and Amazon—has sparked controversy over how AI-enabled infrastructure is integrated into government operations. The debate has not focused solely on legality; it has centred on leverage: once critical systems are embedded, who controls their evolution, safeguards and permissible uses?
In the United Kingdom, recurring debates around Palantir Technologies contracts reveal similar concerns. When analytics platforms become deeply integrated into health services or defence institutions, dependency becomes structural. Even in lawful arrangements, the question of sovereignty lingers.
Across these cases, the pattern is consistent. Artificial intelligence is no longer a peripheral technology; it is becoming a strategic infrastructure whose transformation has global implications. AI companies now compete for state alignment, beyond the usual consumer markets or enterprise clients. While defence contracts offer revenue and prestige and signal trustworthiness and strategic relevance, they also expose firms to political pressure and moral scrutiny.
If American AI companies draw strict military red lines, they may strengthen their credibility among segments of the public wary of autonomous weapons or surveillance overreach. At the same time, they risk appearing unreliable to defence institutions seeking flexible tools. Governments may respond by favouring competitors perceived as more aligned—or by investing directly in domestic public-sector AI capabilities to reduce reliance on private vendors.
Global competition intensifies this dynamic. Chinese AI firms operate within a framework where state integration is expected. European firms navigate a regulatory environment shaped by precautionary norms and human rights language. If US firms appear reluctant to fully support national security use cases, policymakers may reassess the industrial strategy surrounding AI.
Corporate safety policies thus intersect with geopolitics. They shape export controls, alliance structures and research funding decisions. They influence how governments define “trusted” suppliers.
For smaller states, especially across Africa, Latin America and parts of Asia, the implications are sharper. Most frontier AI infrastructure is foreign-owned, and governments in these regions depend on access to models and cloud systems headquartered elsewhere. Their bargaining power is limited.
If major powers can face sudden procurement disruptions or disputes over terms of service, less powerful states are even more vulnerable. Digital sovereignty in the AI era will depend not only on regulatory frameworks, but on negotiating durable access to externally controlled infrastructure.
The historical contrast with ARPANET is instructive. The early internet was publicly funded and architected with an ethos of distributed resilience. AI’s core capabilities are concentrated, proprietary and computationally expensive. That structural difference matters. It means sovereignty questions surface earlier and more sharply.
Looking ahead, several trajectories are plausible as governments may seek contractual overriding provisions that limit corporate discretion in defined national security contexts. They may expand public funding for domestic AI labs to reduce reliance on private firms. They may use industrial policy to align corporate incentives with state priorities.
READ ALSO: NCAA directs airline to refund passengers over VAT charges
Companies, meanwhile, may formalise constitutional-style governance structures—independent oversight boards, published red-line commitments, structured engagement with defence agencies—to maintain legitimacy while continuing to serve governments. Some may differentiate offerings, separating civilian and defence-oriented deployments under negotiated frameworks.
The risk is fragmentation. If disputes escalate into blacklists and retaliatory measures, AI ecosystems could split along geopolitical lines, standards may diverge and access to frontier systems may become conditioned on alliance membership. The global AI market could harden into blocs.
There is also a democratic dimension that should not be overlooked. Citizens have legitimate concerns about how AI is used in military and intelligence settings. They also expect governments to maintain security and strategic capacity. The tension between corporate safeguards and sovereign authority should not be resolved through executive fiat or corporate unilateralism alone. It requires institutional clarity and oversight.
The confrontation between the Pentagon and Anthropic is not the endpoint of this struggle. It is an early inflexion point. It signals that advanced AI systems have crossed a threshold. They are no longer optional enhancements. They are part of the machinery of statecraft.
The internet began as a national security project and evolved into a global infrastructure shaped by multiple actors. Artificial intelligence is emerging as a global infrastructure first, before states have fully settled its rules.
The question is not whether sovereignty will reassert itself. It will. The question is how.
If governments seek absolute control, innovation may constrict, and trust may erode. If corporations claim moral authority without democratic accountability, public legitimacy will suffer.
The balance between them will define not only the future of AI governance but the distribution of power in the decades ahead.
What is at stake is not a single contract. It is the architecture of authority in a world where intelligence itself has become infrastructure.
