Abstract
This article, the second in a series on why artificial intelligence (AI) is unlike any other technology, explores three distinctive features of AI: its unprecedented manipulation potential, cognitive decline risks and power concentration, and knowledge privatisation.
This study traces the evolution of propaganda from mass media to personalised social media bubbles, highlighting AI’s capability to tailor persuasive messages with extreme granularity through big data integration, and raises ethical concerns about privacy and democratic integrity.
This article warns of cognitive decline due to overreliance on AI-generated summaries, which undermines critical thinking and may lead to societal decline, supported by recent academic studies.
Additionally, it discusses the role of private AI companies, such as Anthropic, as influential non-state actors in the military and security domains, contributing to the privatisation of information and power, and challenging traditional state monopolies on violence and accountability. The complex implications of AI’s societal integration call for careful scrutiny of its ethical, cognitive, and political impacts.
Introduction
The first part of the miniseries under the title: “Why isn’t artificial intelligence just another technology?” (available here) focused on two features that arguably make AI technology unlike any other in human history: the black-box phenomenon and emergent behaviour. This article explores three distinct features of AI: manipulation potential, cognitive decline, and the privatisation of knowledge (power).
3. Manipulation – How do we know what is true anymore?
AI presents, in the author’s view, unprecedented manipulative potential. To begin with, scholars and society leaders are aware of the manipulation capacity (shaping and framing) of mass media. Mass media have historically been used to “shape” public opinion through organised propaganda, especially in wartime and by political and economic elites. During World War I, governments systematically employed newspapers, posters, and films to mobilise populations and demonise enemies. Harold Lasswell offered an early systematic analysis, defining propaganda as managing collective attitudes through the manipulation of significant symbols.[1]
As for the interwar years, Walter Lippmann argued that citizens form opinions from simplified pictures in their heads supplied by the press rather than from direct experience, enabling elites to manufacture consent.[2] Edward Bernays applied psychological principles to public relations, advocating the engineering of consent via mass media so that leaders could influence public habits and political attitudes without the audience fully recognising the process.[3]
World War II intensified these practices. Nazi authorities centralised radio, cinema, and print under Joseph Goebbels (Reich Minister of Propaganda from 1933 until 1945) to disseminate ideology and suppress dissent, while Allied governments coordinated comparable campaigns for recruitment and morale.[4] After the war, empirical research identified subtler mechanisms. Scholars demonstrated agenda-setting: the issues given prominence in news coverage become those the public regards as most important.[5] Herman and Chomsky proposed a structural propaganda model in which ownership concentration, advertising dependence, official sourcing, organised flak, and dominant ideology function as filters that systematically align mainstream media output with elite interests, producing consent without overt coercion.[6] Together, these studies show that mass media have repeatedly served not as neutral conduits but as instruments for constructing political realities and securing public acquiescence.
Then came social media. Social media is personalised, and this arguably makes all the difference. A social media user lives in a bubble of infotainment (information combined with entertainment), which differs fundamentally from the bubbles created by mass media. Propaganda spread by mass media, at its core, is based on the idea that a group of people (such as a nation) is exposed to the same type of information, misinformation, or even disinformation (often a combination of the three) and therefore shares core perceptions/misperceptions of reality. Social media are individualised, which means that each user inhabits their own cognitively biased version of reality, shaped by algorithms.
Scholars recognise that social-media platforms have been systematically used to manipulate public opinion through computational propaganda, the deployment of algorithms, automation, and curated human activity to spread misleading information at scale. [7] Governments and political parties operate cyber troops that employ bots, troll farms, and coordinated inauthentic accounts to amplify disinformation, fabricate consensus, and drown out dissent during elections and crises. Social bots disproportionately seed and accelerate low-credibility content in its earliest diffusion stages, after which human users further propagate it.[8] Platform algorithms that optimise for engagement simultaneously privilege emotionally charged and polarising material, producing asymmetric information environments that radicalise audiences and erode shared factual baselines.[9] Collectively, these practices distort agenda-setting, polarise publics, and undermine democratic deliberation by making manufactured narratives appear organically popular.
Arguably, AI technology brings even greater potential for manipulation. The level of granularity (detail, scale, or precision in a set of data, a system, or an analysis) and responsiveness is unparalleled historically. AI’s integration with big data enables the creation of detailed individual profiles, which can be used to influence opinions and behaviours. This is evident in political campaigns, where micro-targeting and personalised messaging have become standard practice.[10] Such practices raise ethical concerns, as they can undermine democratic processes and individual autonomy.
AI significantly enhances both propaganda and surveillance by introducing personalisation, scalability, and real-time adaptability. For example, AI-driven propaganda can tailor messages to individual preferences and beliefs, thereby increasing their persuasive power.[11] Similarly, AI surveillance systems can simultaneously monitor vast populations, identifying and responding to potential threats with unprecedented efficiency.[12]
The use of AI in propaganda and surveillance raises ethical concerns because it can erode privacy, manipulate opinions, and undermine democratic processes. These concerns are exacerbated by the lack of transparency in AI algorithms and the potential for bias in decision-making.[13]
In a recent experiment by The Economist, it was found that AI models are peculiarly biased, especially when it comes to political views or, more generally, philosophical outlook. Consequently, they tend to be more secular and more liberal than an average Western citizen (this bias naturally refers to “Western AI” as opposed to Chinese AI models, which have an inbuilt socialist core values bias).[14] This already presents a growing problem for individuals who increasingly form their understanding of the world with the help of AI, especially when they ask AI for personal advice on major life decisions, such as political voting or life coaching. This creates a societal AI dependence that will certainly bear tangible fruits in the future if it is not bearing them already – namely cognitive disorientation or, in a worst-case scenario, cognitive decline. A highly AI-dependent society becomes fair game for all sorts of intended and unintended manipulation practices.
4. Cognitive decline – the disappearing art of independent thinking
“Once men turned their thinking over to machines in the hope that this would set them free. But that only permitted other men with machines to enslave them.” This famous quote comes from Frank Herbert’s 1965 science fiction masterpiece, Dune. It serves as a cautionary tale against overreliance on technology, warning that delegating thought to machines allows power and control to concentrate.
Arguably, humanity stands at a turning point. Unlike any other technology humans have created, AI has the potential to undermine, if not destroy, our critical thinking skills. I believe several factors explain this unprecedented situation. Two of them stand out as particularly obvious.
First, we live in times of an overabundance of all sorts. Much of it is actually partly untrue or presented in a way to create an untrue message (misinformation), yet a considerable amount is also factually wrong (disinformation). This is especially important in knowledge-driven societies, which have turned information and knowledge into the next currency of power.
In his phenomenal book titled The Third Wave, written 46 years ago by Alvin Toffler argued that industrial society rested on a tightly coupled “info-sphere” of mass media, post offices and telephones that delivered standardised messages to standardised populations.[15] That system is now overloaded. Diversity of energy sources, production methods, family forms and values multiplies the information that must circulate if society is to hold together. Computers and telecommunications therefore generate an explosive new info-sphere: interactive, de-massified media, machine-to-machine traffic and an “intelligent environment.” Social memory, once frozen in print and archives, becomes both vast and active; data can be processed, not merely stored.
This shift dismantles Second-Wave principles—standardisation, centralisation, and synchronisation. Mass production yields to customised, information-rich goods; mass media splinter into targeted, two-way channels that “de-massify the mind.” Power migrates from those who control factories or money to those who control knowledge. Work relocates toward the “electronic cottage,” where people become “prosumers,” simultaneously producing and consuming information and goods.
The result is not merely more data, but a new social architecture: decentralised, minority-powered, faster, and more fragmented. Toffler warned of “blip culture” and information overload, even as he saw the Third Wave as the most creative restructuring in history, provided societies learn to synthesise rather than drown in the flood of data.
The direct consequence of information overabundance is that our brains often feel confused, if not lost. In other words, the human brain was not designed to process such a large amount of information in such a short time. However, because it was designed to optimise performance, when AI summary functionality is offered, our brains do not hesitate. We subconsciously and quite automatically reach out for the AI summary.
The second factor that makes AI technology particularly attractive for an average information consumer, perhaps even for policymakers who always work under the pressure of time, is speed. We live in an era of unprecedented rush and therefore instantaneity. Living and being evaluated round the clock and by the second, the human race has learned not to waste valuable time – “time is money,” we are told by neoclassical liberals. Every second counts. Consequently, we need answers to questions we have not even properly formed in our heads, and we need them now; better, we need them for yesterday.
This constant pressure to be informed and opinionated is self-inflicted, but we kowtow to it with little reflection. We have become information-consuming and information-producing (information-prosuming) automatons. AI technology offers an unprecedented tool that allows us, as information-prosuming automatons,[16] to navigate our busy lives under the constant pressure of time and impossible-to-meet demands of our supervisors. (The trend is identifiable all too well, even in academia, which should, by definition, focus on a slow and in-depth approach).
Naturally, we resort to AI and its summaries. Instead of spending hours reading the source materials and critically evaluating them before forming our own understanding, we limit ourselves to an AI summary, even internet browsers now offer an AI summary first before presenting the user with a list of primary sources.
The effects are already visible and, sadly, very negative. Academics leave no room for doubt. In a typical research paper recently published by an international couple of scholars in Computers in Human Behaviour Reports, Noorbehbahani and Oyibo claim as follows:
”The review reveals that AI-overdependence poses a significant threat to higher-order cognitive skills, especially critical thinking, complex decision-making, and analytical reasoning, while also undermining creativity, memory, and executive functions. These effects stem from well-documented mechanisms, such as cognitive offloading, metacognitive laziness, and automation bias, compounded by the perceived authority and convenience of AI systems. Vulnerabilities are most pronounced among younger learners, individuals with lower educational attainment, and professionals in high-stakes fields such as medicine, where a levelling effect risks deskilling both novices and experts. Moreover, the broader societal implications, from passive learning and professional deskilling to digital inequities and threats to academic integrity, underscore the need for a holistic mitigation strategy that integrates educational initiatives, human-centred AI design, and robust policy frameworks to ensure AI augments rather than supplants human cognition.” [17]
Overreliance on AI-generated content is potentially disastrous not only because we are depriving ourselves of our greatest strength and weapon – critical thought, but also because, as other scholars recently proved, AI is becoming less and less ”intelligent” itself – at least the popularly used large Language Models (LLMs) such as ChatGPT. A recent paper published in none other than Nature claims that “indiscriminately learning from data produced by other models causes ‘model collapse’—a degenerative process whereby, over time, models forget the true underlying data distribution, even in the absence of a shift in the distribution over time.” [18] In other words, because much of the information available online has been produced by AI (mostly LLMs such as GPT-3), these models are increasingly trained on information already produced by AI. This effect leads to a decline in quality or even a possible future collapse of AI models in general.[19]
In simple, cruel words, unlike any other technology, AI has the potential to dumb down humankind for generations to come.
5. AI companies: Non-state actors who privatise information and power.
AI developers are predominantly private entities. Below, the reader will find a list of these companies and their origins, as well as their major products.
* Generated by Grok on 29.09.2026. Prompt: Major companies are developing AI.
An example is always better than a long lecture; let us just look at one of those companies – Anthropic. Anthropic is a San Francisco AI company founded in 2021 by former OpenAI researchers, including siblings Dario and Daniela Amodei. It builds the Claude family of large language models and related tools such as Claude Code, with a public emphasis on safety, interpretability, and systems that are easier to steer. Enterprises use its models for writing, analysis, coding, and agent-style work, and they are available through Anthropic’s own products as well as major clouds including Amazon and Google.
The company is privately held and has become one of the world’s most valuable AI labs, having raised large rounds from investors, including Amazon and Google. It competes directly with OpenAI, Google DeepMind, and others at the frontier of model capability, while arguing that advanced AI should be developed with stronger safeguards than the industry’s default pace. According to its own publicity, as of 2026, it remains independent rather than a division of a big tech platform, even as those platforms both invest in it and distribute its models.[20]
Anthropic is probably better described as a frontier-model vendor to the U.S. Department of Defense (DoD) than as a traditional prime contractor such as Lockheed Martin or Raytheon. In July 2025 the DoD Chief Digital and Artificial Intelligence Office awarded Anthropic, Google, OpenAI, and xAI prototype agreements with ceilings of $200 million each to develop “agentic AI workflows” for national-security missions.[21] Anthropic also supplied Claude through Palantir and classified cloud partners, becoming the first frontier lab on classified networks and fielding custom “Claude Gov” models for intelligence analysis, planning, cyber work, and simulation.[22]
These deployments sat inside a broader Pentagon turn toward commercial large language models for decision support, including Maven-related analytics and IL-6 (specialised security standard established by the U.S. Department of Defence (DoD) and governed by the Defence Information Systems Agency (DISA) Cloud Computing Security Requirements Guide (SRG) to store and process classified national security information up to the Secret level)[23] hosting of Claude via Amazon Web Services and Palantir stacks.[24] Court declarations state that the Defence Counterintelligence and Security Agency granted Anthropic a top-secret facility clearance and that Claude was authorised for FedRAMP (Federal Risk and Authorisation Management Program), a U.S. government-wide program that provides a standardised approach to security assessment, authorisation, and continuous monitoring for cloud products and services.[25] and DoD Impact Levels 4, and 5.[26] Contractors used Claude Gov as a subcomponent of larger defence systems, which is the sense in which Anthropic functioned as a subcontractor rather than a weapons prime.[27]
The relationship broke in early 2026. Anthropic would support military use but refused “all lawful uses,” including mass domestic surveillance and fully autonomous lethal weapons, citing reliability limits.[28]
Much has been recently published on the importance of AI development for military purposes. It can be said that companies such as Anthropic are major stakeholders in the new military-industrial complex. As such, this fits perfectly into the larger picture known to experts as the privatisation of security, defence, and war. The privatisation of war[29] matters for several reasons. Let us mention the three most important ones: the erosion of state monopoly, using private contractors undermines a government’s exclusive legal claim to use force. Accountability gaps: private contractors often operate in regulatory grey zones, creating hurdles for legal prosecution during abuses. Blurred lines: contractors complicate international humanitarian law, as their status shifts between protected civilians and combatants depending on their actions.[30]
Combined with the “fog of war” phenomenon explored by Carl von Clausewitz in his seminal work On War (Vom Kriege), AI companies offer a solution to the problem AI actively contributes to creating: the overabundance of information. The challenge, however, lies in the fundamental legitimacy of organised violence and security policies, which are traditionally reserved for states that represent public interests and work towards the realisation of public goals rather than private maximisation of economic and political profit.
Conclusion
This paper (the second in a series) argues that artificial intelligence (AI) is not just another technology but a transformative force with unique and profound implications. It emphasises AI’s unprecedented potential for manipulation through hyper-personalised propaganda and surveillance, raising ethical concerns regarding privacy and democratic integrity. This study also warns of cognitive decline due to AI overdependence, which undermines critical thinking and intellectual autonomy and could lead to societal disorientation and intellectual deterioration. Furthermore, it discusses the unprecedented concentration of power in private AI companies, exemplified by firms such as Anthropic, which blur the lines between state and private interests, especially in military contexts. This privatisation of knowledge and power challenges traditional state sovereignty and accountability in terms of security and warfare. Ultimately, this study urges careful consideration of AI’s societal impact of AI, advocating for strategies to mitigate risks while harnessing AI’s benefits. Without critical reflection and regulation, AI threatens to reshape humanity’s cognitive capacities and political landscape in ways never seen before.
References
