One of the most consequential geopolitical and technological races underway is the competition to shape the future of large language models. For a moment, it looked like a race to build one dominant cognitive operating system for humanity. But that is not what the next five to ten years will look like. Three forces will define the LLM landscape of the next decade: fragmentation across countries and cultures, the shift from chatbots to autonomous agents, and a quiet transformation in how each of us receives, interprets and shares information.
Search engines organize our information. Social media channels figure out how to grab and keep our attention. Large language models now shape our interpretation.
The first two layers concentrated power. The LLM layer, by contrast, is decentralizing along political, cultural and commercial lines.
Almost overnight, LLMs have become the front door to knowledge. Increasingly, they do not simply retrieve information; they interpret it for us. We consult them as experts, rely on them as filters for decision-making, and use them to help make sense of our world. In the process, we are outsourcing judgement to machines we have never met and never will.
This should not surprise us. Humans naturally apply social rules and expectations to computers – a phenomenon described by Byron Reeves and Clifford Nass in their “Computers are Social Actors” (CASA) framework in The Media Equation (1996). If a machine can communicate fluently, express emotion and simulate empathy, our social instincts engage almost automatically. That tendency will become far more consequential as LLMs continue to evolve.
Fragmentation — many models, many worldviews
Every LLM embeds assumptions.The key question for any model is not whether it is biased. It is “what are its biases and how transparent are they?” Each LLM can embed its own historical framing, level of censorship, moral assumptions, geopolitical narratives and definitions of what is acceptable. There is no globally accepted governance framework that consistently defines these boundaries across models. Rather, it can be different for each LLM today.
This challenge becomes even more complex across languages.Biases in Hindi, Mandarin, Arabic, Portuguese, Bahasa Indonesia, Russian and Spanish may receive far less international scrutiny than English-language outputs.The world may therefore experience not one AI ecosystem, but several competing cognitive ecosystems.
Fragmentation and Sovereign AI
A major structural shift underway is the rise of sovereign AI.
Countries increasingly want domestic models, local compute infrastructure, regulatory control, cultural alignment, and strategic independence.
China already operates a distinct AI sphere through systems such as DeepSeek, Qwen, ERNIE, and Hunyuan. India is pursuing Sarvam and Indus. France backs Mistral.Canada’s Cohere and Germany’s Aleph Alpha are in a planned merger to create a transatlantic sovereign AI vendor. UAE has Falcon and Jais through TII and G42. Singapore’s AI Singapore program backs SEA-LION, a national open-source LLM family. Saudi Arabia’s Public Investment Fund backs HUMAIN, a sovereign AI company focused on Arabic-language models.
It is logical that each of these LLMs will be influenced by language, regulation, compute access, procurement ecosystems, and cognitive alignment.
Open-weight models and asymmetric power
Another major development is the rapid spread of open-weight models.Techniques such as Low-Rank Adaptation (LoRA) allow organizations or individuals to fine-tune powerful models cheaply and quickly. Models can be modified for specialized capability, ideological alignment, style adaptation, or the removal of alignment and safety constraints.
Many open-weight ecosystems contain uncensored variants, often available on platforms, such as Hugging Face, a central hub for open-source AI models. This creates a strategic asymmetry. Advanced AI capabilities are no longer confined to major state actors or frontier labs. Adversaries, extremist groups, criminal organizations and foreign influence operations increasingly have access to highly capable systems.
The Rise of Agentic Systems
While the world fragments into competing models, a second transformation is changing what those models actually do. Today, we still think about AI as chatbots, but that framing is already becoming outdated. LLMs are evolving into agentic systems that call APIs, execute code, coordinate workflows, verify outputs, and operate semi-autonomously. In practical terms, agents will book the travel, draft the contract, monitor the competitor, screen the resumes, reconcile the invoices, prepare the briefing and flag what changed overnight — often calling other agents along the way.
Within five years, much of the information arriving at our desks will likely have been gathered, filtered and summarized by an agent before we read a word of it.The interface shifts from “asking questions” to “delegating objectives.” In this sense, the LLM itself disappears into the background — much like relational databases disappeared into modern computing infrastructure.
The Battle Over Cognitive Infrastructure
Put these two forces together and the picture changes for every leader, every citizen, every reader.
How we receive information. Each of us will increasingly see the world through whichever LLM sits between us and it. That model carries its own training data, its own guardrails, its own omissions. Two colleagues asking the same question of two different systems may get two materially different answers — and neither will know what was left out.
How we interpret information. Agents will not deliver raw material. They will deliver conclusions, summaries, and recommendations. The intermediate steps — the sources weighed, the alternatives discarded — will happen out of sight. We will be tempted to accept what arrives, because the cost of checking will be high and the appearance of competence will be persuasive.
How we share information. Increasingly, the message I send is drafted by my agent and read by yours. Provenance gets murky. Tone gets averaged. Persuasion runs through systems neither of us fully controls. Citizens can gradually lose trust in institutions, experts and media altogether – and societies with weakened shared trust become far more vulnerable to manipulation, polarization and coercion.
For intelligence services, this represents a shift in who controls the collection, preprocessing and interpretation layers that sit between raw data and national-level judgement.
What this asks of us
The United States currently retains major advantages (frontier research, semiconductor ecosystems, hyperscale cloud infrastructure, venture capital, and global platform reach), but the strategic environment is changing quickly. American developers increasingly use Chinese open-weight models because of cost-performance advantages. Open-weight models are publicly available, allowing anyone to run, modify, fine-tune or adapt them to their liking. The visible layer of perhaps dozens of major frontier models understates the true landscape. The real surface area lies in the derivatives, adapters and localized systems proliferating worldwide. The battle over AI and LLMs is not simply about economic advantage or technology leadership. It is about who will shape the cognitive architecture through which billions of people understand truth, authority, identity and reality.
The defining question of the next decade may be “Which system do we collectively trust — and what do we still insist on judging for ourselves”. Because whichever systems mediate knowledge, memory, interpretation, persuasion, and trust will increasingly shape the operating system of human society itself. The good news is the infrastructure is being built, the rules and guidelines are yet to be formalized, governance is an emerging topic and major consolidation has not yet taken place. Our future depends on who preserves human judgement, freedom and trust as our world is transformed by technological advance.
The Cipher Brief is committed to publishing a range of perspectives on national security issues submitted by deeply experienced national security professionals. Opinions expressed are those of the author and do not represent the views or opinions of The Cipher Brief.
Have a perspective to share based on your experience in the national security field? Send it to [email protected] for publication consideration.
Read more expert-driven national security insights, perspective and analysis in The Cipher Brief
