In early April this year, while Pakistan and Oman brokered a ceasefire between the United States, Israel and Iran — a task the UN Security Council had proved unable to perform — a quieter but no less consequential power demonstration was unfolding in Silicon Valley. Anthropic, a San Francisco-based artificial intelligence (AI) company, had already disclosed its most powerful model to date, Mythos, under a controlled access programme called Project Glasswing. The 11 organisations it named as partners were all American. The sole non-American government to receive independent access was Britain’s AI Security Institute. The rest of the world was told to wait.
The juxtaposition is not incidental. The same week that ceasefire diplomacy in Tehran was being brokered by actors Washington treats as peripheral, the world’s most consequential AI company was demonstrating the same geopolitical logic in a different register: the United States sets the terms, and others adapt. Meanwhile, Palantir — the data infrastructure company founded by Peter Thiel and led by Alexander Karp — was distributing a 22-point ideological summary of Karp’s book, “The Technological Republic,” on social media, and its Maven Smart System AI platform had, by its own account, processed 1,000 targeting recommendations within the first 24 hours of Operation Epic Fury.
For the Global South, the Mythos episode and the Palantir manifesto are not separate stories. They are two chapters of the same text: a vision of global order in which technological sovereignty belongs exclusively to those who already hold military and financial sovereignty — and in which everyone else is, at best, a beneficiary, and at worst, a target.
The lock on the door
Mythos is, by Anthropic’s own account, a model that has identified thousands of previously unknown software vulnerabilities — many of them decades old — in systems that run banks, electricity grids and government networks. The company deployed it through Project Glasswing with a small group of technology infrastructure providers, asking them to develop patches before adversaries could exploit the same flaws. The logic is defensive. Anthropic has said it expects comparable models from other organisations within 18 months. The patching window is narrow.
But the execution of this strategy has produced a structural asymmetry. The same banks, electricity grids and government systems carrying these vulnerabilities exist in Jakarta and Nairobi and Mumbai as much as in London and New York. Anthropic has not extended access to those nations. It has extended access to Amazon, Apple and Microsoft.
Britain’s AI Security Institute confirmed in an independent assessment that Mythos can carry out complex cyber operations that no previous model has performed. Germany’s Federal Office for Information Security received no access to the model, only a series of briefings in San Francisco. The European Commission held at least three meetings with Anthropic and could not agree on terms. For the countries of the Global South, there were no meetings at all.
Eduardo Levy Yeyati, a former chief economist at the Central Bank of Argentina and a regional adviser on AI at the Inter-American Development Bank, has described this as a policy wake-up call. His point is well made. But waking up to a problem is different from having the power to solve it. The Global South is not facing a policy gap. It is facing a structural condition — one that has been constructed deliberately, piece by piece, over several years.
The ideology in the code
To understand why access to Mythos was distributed as it was, it is necessary to grasp how Dario Amodei, Anthropic’s chief executive, conceptualises the purpose of powerful AI. His position is not hidden. At the Paris AI Action Summit in February 2025, Mr Amodei argued that democracies must maintain the lead in AI development. He has described the Communist Party of China as the greatest threat in this regard and called for an “entente” strategy — a coalition of democracies that uses AI to achieve decisive strategic and military advantage over adversaries, then distributes the benefits among cooperating democratic nations.
This is not a marginal view within the American tech industry. It is the organising principle of Anthropic’s geopolitical posture. In his January 2026 essay “The Adolescence of Technology,” Mr Amodei elaborated on his concern that powerful AI in the hands of autocratic regimes could produce what he called a “global totalitarian dictatorship.” He identifies autocracies as the primary threat and democracies as the legitimate custodians of transformative technology. The Mythos access list — American companies, one British government body, no Global South representation — is the operational expression of that ideology.
When a private corporation’s ideological framework becomes the de facto governance structure for a technology with global security implications, the distinction between a company and a state begins to dissolve. Anthropic is not elected. It is not treaty-bound. It faces no international accountability mechanism. Yet its decisions about who may access Mythos carry the force of policy for every nation that lacks the infrastructure to build a comparable model itself.
This is what makes the framing of democracies versus autocracies so consequential for the Global South. The binary is not neutral. It is constructed from within a specific political tradition and applied by actors who have historically used the language of democracy to justify actions that democratic majorities around the world have rejected. The invasion of Iraq was presented as the defence of liberal values. Operation Epic Fury, launched on February 28th 2026 against Iran — killing Supreme Leader Sayeed Ali Khamenei in its opening hours, devastating the country’s military and economic infrastructure, and temporarily closing the Strait of Hormuz through which 20% of the world’s oil passes — was described by Washington as the exercise of collective self-defence.
A ceasefire took effect on April 8th 2026, after 38 days of operations and thousands of deaths across Iran, Lebanon, Israel and the Gulf Arab states. Iran’s government estimated its economic damage at between $300bn and $1 trillion. The negotiations that produced the ceasefire were mediated by Pakistan and Oman. The Global South watched, condemned and was overruled.
Those who believe that the same Western powers now determining AI access will treat the Global South differently in the technology domain than they have treated it in the military and energy domains are making an argument that recent history does not support.
Technological Republic and who it excludes
The Anthropic worldview does not exist in isolation. It sits alongside — and in critical ways, complements — an older and more explicit strain of American tech-imperial thinking, which found its latest articulation in the 22-point social media summary that Palantir published this month of Alexander Karp’s book, “The Technological Republic.”
Because we get asked a lot.
— Palantir (@PalantirTech) April 18, 2026
The Technological Republic, in brief.
1. Silicon Valley owes a moral debt to the country that made its rise possible. The engineering elite of Silicon Valley has an affirmative obligation to participate in the defense of the nation.
2. We must rebel…
Palantir is not a political party. It is not a think tank. It is a defence and intelligence contractor that builds the data infrastructure used by militaries, intelligence agencies and law enforcement agencies across the United States and its allies. When such a company begins to speak in sweeping ideological terms about the direction society should take, the relevant question is not whether its ideas are interesting. It is what those ideas justify — and who bears the cost.
The summary’s fifth point is the most revealing in this regard: the question is not whether AI weapons will be built, Mr Karp argues, but who will build them and for what purpose. This is the foundational move of the entire framework. By asserting that AI weaponisation is inevitable, the argument pre-empts any principled objection to participating in it. The tech company becomes not a war-profiteer but a responsible actor who, unlike squeamish rivals, has accepted the burden of building weapons for the right side.
The logic requires a prior agreement that there is a right side. Point 13 of the summary supplies it: the United States, Mr Karp’s text asserts, has advanced progressive values more than any other country in history and offers more opportunity than any other nation to those who are not hereditary elites. Point 14 reinforces it: American power has, Mr Karp argues, ensured nearly a century of peace — a claim that requires considerable selectivity about what constitutes peace, given the Korean War, the Vietnam War, the invasion of Iraq, the drone campaigns across Pakistan and Somalia, and, most recently, the 38-day air campaign against Iran that Palantir’s own Maven Smart System helped execute.
Point 21 states the corollary most bluntly. Some cultures, the summary asserts, have produced vital advances; others remain dysfunctional and regressive. The sentence sits there, unattributed to any specific tradition or nation, but its function in the argument’s architecture is clear: it establishes a hierarchy of civilisations that gives content to the distinction between those who should build AI weapons and those who should bear their effects. It is not a new idea. It is a very old one that, in earlier centuries, provided justification for the colonial enterprise.
Point 22 closes the circle, resisting what Mr Karp calls the shallow temptation of a vacant and hollow pluralism. For societies whose entire political project has consisted of claiming a place in a multilateral order that the West has repeatedly structured to exclude them, this sentence is not an abstraction. It is a direct statement about whether their voice counts.
Kill chain made efficient
The distance between Mr Karp’s ideological framework and its material application is shorter than any comfortable reading of this episode would suggest. Maven Smart System — Palantir’s AI platform, which fuses data from across military intelligence networks to compress targeting decisions from hours to minutes — was, according to the Pentagon’s Chief Digital and AI Officer, central to Operation Epic Fury‘s execution. According to news reports, the system enabled the processing of 1,000 targets within the first 24 hours of strikes. Over three weeks, between 5,500 and 6,000 targets in Iran were struck.
The Maven contract, initially worth $480m when awarded in 2024, had its ceiling raised to $1.3bn in May 2025, alongside a $10bn Army enterprise framework agreement. NATO acquired a version of the platform in March 2025. Maven now has more than 20,000 active military users — a figure that has quadrupled since early 2024.
It is also worth noting, as Reuters has previously reported, that Maven’s operation incorporates AI software from Anthropic — the same company whose Mythos model is now being deployed defensively to scan for vulnerabilities in global infrastructure. The two companies are not separate pillars of American tech power. They are, in key operational respects, connected. Anthropic builds the underlying model capability; Palantir builds the targeting infrastructure that runs on top of it; the US Department of Defense funds and deploys the combined system; and the Global South sits outside all of it, exposed to its effects.
Mr Karp, in the aftermath of Operation Epic Fury, described AI precision targeting as having fundamentally shifted modern warfare. He is right. The question that the 22-point summary does not ask is: shifted it for whom, and in whose favour?
Chips, chokepoints and competition that wasn’t
The semiconductor dimension of this story is, if anything, more explicit than the ideological one. Since October 2022, successive US administrations have imposed a series of export controls designed to prevent China from acquiring the advanced chips necessary to train frontier AI models. The restrictions were tightened in 2023 and 2024, extended to allied nations through diplomatic pressure, and, in March 2025, supplemented by additional blacklisting of Chinese entities.
The effect has been measurable. Huawei produced only 200,000 AI chips in 2025, against the millions produced by US manufacturers. Chinese AI researchers have privately expressed concern that the gap is widening precisely as the strategic consequences of falling behind become clear. The Carnegie Endowment for International Peace’s Matt Sheehan has described Mythos as a second major wake-up call for China’s AI community, noting that chip policy has materially extended the American lead.
The Trump administration has since partially reversed course. In January 2026, the Bureau of Industry and Security formalised a policy permitting exports of Nvidia H200-equivalent chips to China on a case-by-case basis, subject to a 25% revenue-sharing arrangement with the US government — an arrangement whose legality has been questioned by scholars who argue it functions as an unconstitutional tax on export licences. The policy is, in any event, restrictive in practice: it applies only to approved customers, requires detailed end-user certification and excludes the most advanced chips entirely.
The argument that these restrictions protect global security deserves scrutiny. China’s banks, energy infrastructure, and government systems run on the same software that Mythos found vulnerabilities in. China has no access to Mythos. Its developers cannot build models capable of the same level of defensive scanning because they cannot legally obtain the chips needed to train them. If the genuine objective of Western AI policy were global cybersecurity, this arrangement would be difficult to defend. If the objective is to ensure that the United States and its allies hold a structural advantage in the technologies that will define the next generation of economic and military power, the arrangement is entirely coherent.
That coherence is not lost on the nations of the Global South. India, with its large and technically capable AI research community, has watched the Mythos episode from the outside. The Association of Southeast Asian Nations, the African Union and the Gulf Cooperation Council—none of them were named in Project Glasswing’s partner list. They were not consulted about access criteria. They were not offered briefings comparable to those extended to Germany’s Federal Office for Information Security. Levy Yeyati has put the stakes plainly: the ability of a company to unilaterally restrict access to frontier AI, using opaque and unappealable criteria, should concern every government that lacks its own frontier model programme.
World sans treaty
There is no Nuclear Non-Proliferation Treaty for AI. There is no International Atomic Energy Agency to conduct inspections, no agreed protocol for handling the disclosure of a model with Mythos’s cyber capabilities. The closest analogue to a multilateral safety process — the AI Safety Institutes established in Britain, the United States and a handful of other countries — is funded by Western governments and answerable to them.
The Paris AI Action Summit in February 2025 produced no binding commitments. Mr Amodei called it a missed opportunity. The US vice-president used his presence to criticise European AI regulation as an obstacle to innovation. The gap between the pace of AI development and the pace of governance has not narrowed since.
What has changed is the materiality of the stakes. Mythos is not a research paper or a benchmark result. It is a deployed system that has, in weeks, identified security flaws in the infrastructure underpinning the global economy. The organisations working to patch those flaws are American. The timeline for patching — before comparable models emerge elsewhere — is, by Anthropic’s own estimate, 18 months. Nations outside that process are not merely behind: they are exposed.
Mr Karp’s framework, presented as a vision, is in this sense already a description. Palantir’s contracts are signed. Maven has processed tens of thousands of targets. Anthropic has declined to extend access to the Global South. Chip controls have made it structurally difficult for countries outside the American alliance system to build comparable capabilities domestically. The technological republic that Mr Karp describes is not arriving. It is already here. The question that the Global South must now ask — with some urgency — is not whether to engage with this architecture, but on what terms.
Warning
The Mythos episode and the Palantir manifesto together reveal something important about the nature of American AI power: it is not merely technological. It is ideological. It requires that recipient nations accept, at least implicitly, a particular account of which political systems are legitimate, which cultures are capable of managing advanced technology responsibly and which strategic partnerships are permissible. Non-compliance with that account is the price of exclusion — from model access, from chip supply chains and, as Operation Epic Fury demonstrated, from the diplomatic processes that determine when wars begin and end. The Global South has not been invited to design the technological republic. That is, at minimum, a reason to begin designing something else.
East Post is an independent geopolitical analysis portal covering South Asia and global power dynamics for international audiences. Views expressed are analytical and do not constitute endorsement of any state or non-state actor.
Join our channels on Telegram and WhatsApp to receive geopolitical updates, videos and more.
