On 5 January, I chose the word ‘clarity’ to describe AI developments in 2025. At the end of the year, we have more clarity and less AI hype. It is clear that AI is not a magic wand that will solve all of humanity’s problems, as many have argued in previous years. Yet, the AI scene remains foggy, mainly due to the continuation of the AI hype narrative. While 2025 provided us with the contours of AI reality, clarity will increase in 2026 as we face a sharper contrast between, on the one hand, inflated expectations and the value of the AI industry, and, on the other hand, the still-limited use of AI by companies and society. Last year was shaped by two main tech developments that occurred on January 20th, when President Trump was inaugurated, and DeepSeek released an open-source AI reasoning model. The remainder of 2025 was marked by addressing these two developments. Trump’s marriage of convenience with Silicon Valley removed regulatory guardrails for AI, content policy, and cryptocurrency. AI investment helped Trump’s economic revival narrative, as it contributed between 34% and 50% to the rise of US GDP in 2025. A lot of investment came from abroad, especially from oil-rich Gulf countries. It resulted in an inflated AI bubble, which the world will have to cope with in 2026. DeepSeek shook the AI space in three main ways: Firstly, it challenged the simplified formula behind the AI bonanza: more Nvidia GPUs = better AI. Second, DeepSeek’s open-source gambit challenged dominant proprietary players: OpenAI, Anthropic, and DeepMind. Third, it reshaped AI geopolitics by having Chinese companies act as shapers of AI developments through open-source solutions—the approach used by Western actors in establishing Internet, web, and Linux standards. Trump’s presidency had more immediate and visible impacts, while DeepSeek’s ‘moment’ created a structural shift in AI development. Here is the Scorecard of predictions from January 2025: Technology | Geostrategy | Governance | Security | Human Rights | Economy | Standards | Content | Development | Environment Clarity is the keyword for AI and digital developments in 2025. It follows the hype of 2023 and the grounding of 2024. In 2025, we will better understand AI’s risks, opportunities, and policy issues that must be regulated. By clarity, we also mean a return to digital basics. It’s easy to forget that even the most cutting-edge AI is built on decades-old foundations—like the humble TCP/IP protocol that underpins our digital reality. Our 10 forecasts for 2025 begin with the evolution of AI technology itself, exploring how geostrategic interests and positions are shaping its development. From there, we delve into governance, where these interests crystallise into policies and regulations. With the stage set, we turn to key issues: security, human rights, the economy, standards, content, the environment, and global development. The first reality check for our 10 forecasts in 2025 starts on 20 January, when we test whether we were right about predictions for President Trump’s tech priorities, followed by an outlook for the rest of the year. Throughout the year, we will continuously monitor forecasts around the monitoring questions listed below in each section. You can also submit your questions and topics on AI and digitalisation that you want us to monitor in 2025. Wishing you all the best for 2025! Jovan Kurbalija Topics: Technology | Geostrategy | Governance | Security | Human Rights | Economy | Standards | Content | Development | Environment
Be careful with AI predictions and forecasts! Any AI prediction, including this one, should be approached with caution. The history of AI predictions is riddled with inaccuracies. Take Geoffrey Hinton, the 2024 Nobel Prize Laureate, who declared back in 2016: ‘We should stop training radiologists now. It’s completely obvious that within five years, deep learning will outperform radiologists.‘ Yet, radiology—like many other professions—remains alive and well. The list of flawed AI predictions is extensive, ranging from exaggerated risks of open-source AI to the impact of AI on elections and so on. Latest updates (January – March 2025) Recent AI Predictions and Their Realism Hinton’s false prediction illustrates a common misconception about AI’s capabilities and limitations. Here’s why radiology wasn’t the ‘low-hanging fruit’ for AI that many thought it would be—and what we can learn from this miscalculation: Quality of data: AI thrives on vast amounts of high-quality, annotated data. But in radiology, getting that data is no mean feat. Medical images are sensitive, requiring strict privacy protections and expert labeling. Plus, the diversity of images—based on patient demographics, diseases, and imaging techniques—makes it hard for AI models to generalise effectively. What works in one scenario often fails in another. Lack of ‘ground truth’: Unlike identifying cats or dogs, interpreting medical images is complex. Radiologists often disagree on findings, and images can contain multiple abnormalities that need precise detection and analysis. This lack of a clear ‘ground truth’ makes it tough to evaluate AI’s performance. Workflow woes: Even if an AI model performs well in a lab, integrating it into a real-world radiology workflow is a whole other challenge. Radiologists need trustworthy, explainable, and seamlessly integrated tools in their systems. Add to that ethical, legal, and regulatory hurdles and it’s clear why AI adoption in radiology has been slower than expected. So, what’s the takeaway? Hinton’s false prediction reminds us that AI’s potential is vast, but its path is riddled with complexities. Radiology isn’t disappearing—it’s evolving, with AI as a powerful tool rather than a replacement (read more here). Perhaps the biggest lesson of all is that predicting AI’s impact is about understanding human systems as it is about the technology itself. The future of AI isn’t just about what it can do—it’s about how we choose to use it. Making incorrect predictions might seem harmless, yet it can have real-world consequences. As an example, the overblown fears of existential risks from AI have led to a tsunami of AI governance initiatives, some of which are premature and/or misdirected. We have three suggestions on reviewing predictions about AI developments: Avoid overreacting: Governance and business initiatives should not be based on highly uncertain predictions. The past two years have seen numerous AI governance initiatives driven by speculative risks, which may have diverted resources from more pressing issues. Develop a balanced outlook: A balanced combination of tools and approaches is essential to address various risks while focusing on key ones. As many have come to understand that the emphasis on existential risk was misguided and overinflated in 2023, it is crucial not to swing to the opposite extreme and disregard existential risks entirely. Include non-technical expertise: While technical expertise on how AI functions is important, it must be paired with deep insights into how societies adopt, adapt to, and integrate new technologies. Balancing these perspectives will be key to shaping AI’s role in a way that is both innovative and socially responsible. Diplo’s experience in predicting digital governance trends since 2011 shows that change often occurs much slower than perceived. Areas like data governance and content policies are examples of glacial progress in technology regulation. In hindsight, the ‘slow governance for fast technology’ approach has, despite its challenges, facilitated technological growth. Consult Diplo’s previous predictions for the last 14 years: 2011 | 2012 | 2013 | 2014 | 2015 | 2016 | 2017 | 2018 | 2019 | 2020 | 2022 | 2021 | 2022 | 2023 | 2024 As President Trump prepares to take office on January 20, we can expect a mix of continuity and subtle policy shifts in the tech realm, as outlined here: 🔹 Historical continuity: Trump’s pro-business stance aligns with the long-standing US tradition of private sector-led innovation, resisting international regulations that could constrain American tech businesses. 🔹 Content regulation: Expect a softer approach on combating misinformation, as already signalled by policy changes in Musk’s X and Zuckerberg’s Meta. Key issues will be the future of Section 230, the US law that provides immunity for online platforms from liability for third-party content, and the trend of stricter content regulation worldwide. 🔹 AI policy: Trump will likely scrap Biden’s Executive Order on AI safety, focusing instead on innovation, workforce upskilling, and global competitiveness. This shift will sync with a global shift from AI safety narratives to those of opportunity and development. 🔹 Geostrategy: Relations with China will continue with more export restrictions for advanced technology and the solidifying of the tech block of like-minded countries. 🔹 Digital taxes: Unresolved tensions over tech taxation will resurface after the failure of the OECD to introduce global solutions. Countries like Germany and France will revisit digital tax policies, potentially clashing with US tech giants and the Trump administration. 🔹 Cryptocurrencies: The crypto industry stands to benefit from the expectation of Trump’s introduction of crypto-friendly regulations, including a strategic crypto reserve and improved access to banking services. 🔹 TikTok’s future: Trump will be open to a TikTok deal, avoiding major internal disruptions and a risky precedent that can expose US tech companies to ‘nationalisation’ pressures in other jurisdictions. From bigger is better to smaller is smarter DeepSeek challenged the dominant assumption that more Nvidia GPUs automatically mean better AI. Throughout 2025, each new model demonstrated that LLMs are reaching the scalability plateau, where additional computing power yields only limited advances. AI has entered an era of diminishing returns. This trend coincided with massive continued investment in AI hardware, finally closing the gap between inflated hype and reality. AI Becomes a Commodity Apple Inc. has faced plenty of criticism from Wall Street for not spending as aggressively on artificial intelligence as its Big Tech rivals. But that strategy is suddenly a blessing for the iPhone maker. Investors are beginning to scrutinize the huge sums companies like OpenAI, Meta Platforms Inc. and Microsoft Corp. are spending on AI, leading to heavy volatility in what had been some of the year’s biggest momentum plays. As a result, Apple’s position is being re-evaluated. The Rise of Bottom-Up AI From bigger is better to smaller is smarter In 2025, the bigger is better pattern in AI will be challenged. Smaller models, such as DeepSeek, are increasingly outperforming larger ones as they are much more cost-effective to train and use. For instance, the cost of training DeepSeek is US 5.6 million—1% of the cost of training Claude 3.5 Sonnet or Meta’s Llama 3, and 500 times less than Elon Musk’s Grok, which uses 100,000 Nvidia H100 GPUs. DeepSeek provides better reasoning for a fraction of the cost. For example, inference (answering questions) costs 3% of that of OpenAI. Think of buying an iPhone for USD 30 instead of USD 1000! Additionally, smaller models require less energy for training and operations, making them more environmentally friendly. This shift from ‘bigger is better’ to ‘small is beautiful’ in AI will have several significant impacts: Reality check for the artificial general intelligence (AGI) narrative: Since the launch of ChatGPT in November 2022, there has been widespread speculation about the arrival of AI that can think and act like humans in any field. As AGI is not likely to emerge soon, there is an increasing narrative shift towards AI agents (see below) as a substitute for the envisaged AGI. This reframing of the AGI discussion will accelerate in 2025. The rise of AI agents: An AI agent is a system or program that is capable of autonomously performing tasks on behalf of a user or another system by designing its workflow and utilising available tools, thus encompassing a wide range of functionalities beyond mere natural language processing, including decision-making, problem-solving, interacting with external environments, and executing actions (IBM). A combination of AI, human expertise, and specific use cases will converge around AI agents, which will dominate the AI landscape in 2025. Similar to previous technological phases, there will be a ‘tech hype’ or tsunami of AI agents promising solutions to all our problems. The best way to navigate this hype is to choose AI agents based on practical needs—whether a new tool is nice or genuinely helpful in tasks ranging from drafting diplomatic agreements to arranging protocol details for official dinners or summarising meeting reports. Growth of bottom-up AI: Affordable and smaller AI models will facilitate the growth of bottom-up AI, addressing the specific needs of communities, companies, and countries. This approach is both technically more feasible and financially more viable. Moreover, it grounds increasingly abstract discussions about ethics and biases in local cultural contexts. Local communities can develop AI that reflects their unique cultural and ethical values while protecting data and knowledge in simple yet effective ways. Open-source AI wins: The open-source approach has won the competition against closed models. The 2023 narrative that open-source AI could pose dangers has proven unfounded. The digital world is returning to open AI as the preferred approach. While the AI industry initially pushed for closed models, citing concerns about misuse, open-source AI has emerged as the key approach. In addition to platforms in the USA and Europe, China is becoming a major player in open-source AI. Today, Qwen 2.5 is the most powerful open-source AI model out there, and open-source AI will be established as the dominant approach. Necessity is the mother of AI invention as well: DeepSeek developed one of the most powerful AI models using a fraction of the funding available to major companies and older-generation Nvidia GPUs, which are still exportable to China. These limitations were overcome thanks to the innovative approach and creativity of the development team. AI is becoming a commodity: Advances in technology have made it possible to develop large language models (LLMs) and AI platforms with limited resources. The almost daily development of new LLMs has led to what is described in China as the ‘war of a hundred models’. Affordable AI is growing rapidly worldwide. AI transformation is a complex task: In 2025, many businesses and organisations will search for a formula for making shortcuts in the AI Gartner Curve towards the Plateau of productivity. AI is an affordable commodity, but AI transformation is very ‘expensive’. At first glance, this statement seems paradoxical. How can something so readily available be almost priceless. Yet, it captures today’s dilemma facing businesses, governments, and organisations. AI has become accessible to many—you can create an AI chatbot in hours—but unlocking its potential requires far more than technology. It demands a shift in professional cultures, a break from old routines, and embracing new ways of thinking and problem-solving. Effective AI adoption isn’t about purchasing the latest software or algorithms; it’s about reshaping how we work, collaborate, and innovate. AI transformation requires challenging the status quo, rethinking long-held practices, and fostering a culture of continuous learning and adaptability. But the rewards are immense. The good news? This journey isn’t just about efficiency, profit, or technological shift; it’s a cultural and philosophical evolution of contributing to the well-being of our communities, countries, and humanity as a whole. Moreover, AI nudges us to reflect on what it means to be human—both ontologically and spiritually. As we use AI effectively, we may find ourselves closer to answering some of humanity’s eternal questions of purpose and happiness. Losing grip of the NVIDIA monopoly NVIDIA’s dominance is not as unshakeable as it appeared at the start of the year, due to several key developments. First, computing power is becoming less critical for AI performance compared to proprietary knowledge and data. Second, China has successfully developed domestic GPU technology sufficient for most of its AI needs, reducing its reliance. Third, competitors like Google have unveiled new GPUs that directly challenge NVIDIA’s market hold. Geoeomotions take centar stage of AI developments As the AI revolution shifts from raw computation to broad societal adoption, public attitude and sentiment have become critical factors. This “geo-emotional” divide gained prominence in 2025, with Asian societies showing greater excitement about AI, while the Anglosphere remains more cautious. On opposite ends of the spectrum are China, where 80% of the population is excited about AI developments, and the United States, where only 34% share that enthusiasm. This readiness—or reluctance—to embrace AI will fundamentally shape the transformation of businesses and societies in the years ahead. GEO politics/economy/emotions in the AI era In 2025, geography will play an increasingly significant role in shaping global politics and economies. The strength of national borders as barriers to the flow of capital, people, and goods will intensify. A central question will revolve around the global flow of data. So far, internet traffic has resisted significant fragmentation, but will this remain the case in 2025? As digital geostrategy gains prominence, the influence of China and the USA in the digital realm will grow. However, the geostrategic landscape will not be purely bipolar. In certain sectors, such as digital trade, new power centres are emerging, particularly in the Global South, reflecting a more multipolar digital economy. Geopolitics often dominate media coverage, focusing on the use of technology to advance security and other non-economic interests. Geoeconomics, on the other hand, centres on accumulating wealth and expanding markets. Meanwhile, emotions and perceptions help to explain why societies embrace or resist technology, reflecting a spectrum of enthusiasm, bias, and fear. Together, geopolitics, geoeconomics, and geoemotions shape the complex interplay between technology, society, and global power dynamics in the 21st century. Digital networks and AI developments are critical assets for countries worldwide. Thus, they become central to national security, the projection of power, and the protection of national interests. Over the last few years, political considerations have been prioritised over economic interests. This is particularly noticeable in various export restriction regimes on semiconductors, limited market access, the deployment of submarine cables, and the launching of satellites. The growth of the semiconductor industry will slow down in 2025 to 12.5% growth from 16% in 2024, according to the World Semiconductor Trade Association. New AI models which require less processing power for training and fewer GPUs could result in less demand for GPUs. The USA starts returning the semiconductor industry home with the opening of a fabrication plant (or ‘fab’) in Arizona. As the USA restricted the export of semiconductors, China invested heavily in its local industry. China will focus on producing less advanced but essential, so-called ‘mature-node’ chips that can have wider economic use. Till now, AI has been processed in powerful data centres. In 2025, we will notice the emergence of ‘AI factories,’ purposely developed to train AI models. Nvidia is preparing a new Blackwell GPU, which will be used for AI processing. EVENTS First meeting of the International Advisory Body for Submarine Cable Resilience | late February 2025, Abuja IPCC Plenary Meeting 15-17 April 2025, Montreal In 2025, the trend of ‘securitisation’ of the economy will significantly impact the tech sector, compelling companies to align their economic interests more closely with their home countries’ political and security priorities. This shift is driven by the growing recognition of technology as a strategic asset in geopolitics. Currently, tech companies are wielding unprecedented power, often surpassing the GDPs of entire nations. For instance, Apple’s market capitalisation in January 2025 was US$3.524 trillion, Nvidia’s was US$3.262 trillion, and Microsoft’s was US$3.101 trillion, each of which is comparable to the total 2023 GDP of the entire African continent (US$3.1 trillion) and close to the GDPs of the UK (US$2.27 trillion), France (US$3.03 trillion), and India (US$3.57 trillion). Other tech giants like Amazon, Meta, Alphabet, Alibaba, and Tencent have similar income and profit levels, further emphasising the economic clout of the tech industry. The power of tech companies extends far beyond technology, permeating various aspects of global society and governance, including social influence, data centralisation, and political impact. No company in the history of humanity, including the East India Company, has had such combined power extending beyond the economy to social and political realms. In 2025, tech giants will likely face pushback from national governments and local companies. For example, Flipkart in India, which controls one-third of the market, and Mercado Libre, an Argentinian firm, are challenging global tech giants in their respective regions. According to The Economist, this trend is also supported by bottom-up financing by local financial initiatives such as M-Pesa in Africa and Nubank in Brazil and South America. Furthermore, governments worldwide are increasingly imposing ‘data localisation’ requirements, which will significantly impact the business models of tech companies that rely on the free flow of data across national borders. In summary, 2025 will be a pivotal year for the tech sector as it navigates the dual pressures of aligning with national security priorities, adapting to regulatory changes, facing competition from regional players, and addressing the financial inclusion needs of underserved populations in the Global South. In 2025, geoemotions will impact the acceptance and growth of AI. If societies fear AI, the use of technology will be limited. A 2024 Ipsos study positions countries in two dimensions: nervous and excited about AI.
Generally speaking, societies from the Anglosphere, including the USA, are highly nervous and display low excitement about AI. This view is probably shaped by the strong media campaign about AI risks in 2023. On the opposite side are Asian societies with high excitement and low nervousness about AI. Similar YouGov research from September 2024 showed that the most positive attitude towards AI is in the UAE (60%), while the least enthusiastic is in the USA (17%). In 2025, these trends will likely change as Trump’s administration will shift focus from safety issues towards AI opportunities and narratives. This political shift is likely to impact media and academic coverage of AI. Temporary consensus and the postponement of important decisions The UN negotiations on WSIS+20, which determined the future of the World Summit on the Information Society, successfully produced a consensus report. While any agreement in the current geopolitical climate is a success, the outcome was less a policy breakthrough and more a search for the lowest common denominator. Critically, WSIS+20 failed to resolve the risk of creating two parallel policy processes—the Global Digital Compact (GDC) and WSIS itself—addressing similar issues, leaving the search for a unified solution to future negotiations. A shift in focus: From existential to existing and exclusion Risks Media focus has shifted away from existential risks, as speculative timelines for an AI superintelligence are constantly pushed back. Instead, existing risks—particularly to education and employment—have gained prominence. Meanwhile, the exclusion risk of AI monopolisation by a few corporations is being counteracted by the proliferation of the technology itself. The rise of bottom-up, open-source AI development is diminishing the relevance of such monopolies. In response, entities like OpenAI are pushing back, arguing for the necessity of centralised, proprietary models on grounds of security and geopolitical competition, notably with China. Negotiations to implementations After a year of intense negotiations on AI and digital governance in 2024, 2025 will shift the focus to implementation as the world works to put resolutions, agreements, and treaties into action. This should be a year of clarity and consolidation, with two key themes echoing across the AI and digital realms: The WSIS framework, shaped between 2003 and 2005, has been tested and refined over two decades. The GDC, introduced in 2024, represents a fresh, dynamic approach to digital governance. Metaphorically, WSIS is a marathon, while the GDC is a sprint. The main challenge in 2025 will be to sync the experience and expertise of the WSIS framework with the new energy of the GDC. This alignment is crucial to avoid duplication and ensure that both frameworks complement each other. The WSIS+20 review in 2025 will provide the perfect context for this synchronisation, offering an opportunity to integrate lessons from the past with the urgency of the present. Hanoi: A new toponym in cyber politics For years, ‘Budapest’ has been a key toponym in cybersecurity, named for the Hungarian city where the 2001 Council of Europe Cybercrime Convention was adopted. This October, Hanoi emerged as a new landmark, hosting the signing of the UN Cybercrime Convention. The treaty will enter into force 90 days after its 40th ratification. Following Budapest, the 2001 Council of Europe Cybercrime Convention host, Hanoi, will emerge as the next toponym in cybercrime language. The Vietnamese capital will host the signing ceremony for the new UN Cybercrime Convention, which will remain open for signatures until 31 December 2026. The convention will enter into force 90 days after the 40th ratification. At the same time, the Ad Hoc Committee on Cybercrime decided that it would complete its work on the convention by holding a session in Vienna, lasting up to five days, one year after the convention’s adoption. Since the convention was adopted on 24 December 2024, this follow-up may occur by the end of 2025. During this session, the committee will draft the rules of procedure for the Conference of the States Parties and other rules outlined in Article 57 of the convention. We expect the convention to enter into force in 2025, with 40+ ratifications. Though this might be tricky, there is considerable diplomatic and political momentum, a need to address growing cybercrime, and relatively general satisfaction with the truly global nature of this binding agreement adopted by consensus in the UN (which is not common these days).
2025 will be an important year for cybersecurity negotiations—the mandate of the UN Open-Ended Working Group (OEWG) on the security of and the use of information and communications technologies is ending in July 2025 with its 11th session. What will follow is a new mechanism for dealing with cybersecurity under the UN auspices. Currently, states disagree on the scope of thematic groups in the future mechanism: (a) while some countries insist on keeping traditional pillars of the OEWG agenda (threats, norms, international law, confidence-building measures, and capacity building), (b) others advocate for a more cross-cutting and policy-orientated nature of such groups. There is also uncertainty regarding the modalities of multistakeholder engagement in the future mechanism. Agreements on these issues are key if states want to hit the ground running and not get tangled in red tape at the beginning of the next mechanism. In the OEWG report, which should be adopted by July 2025, there are a few points where we can expect consensus. CBMs are gaining wider support, including establishing the Global Points of Contact (POC) Directory and capacity-building portal. We may also have some implementation checklists for the existing 11 cyber norms. Support for a voluntary fund for capacity building is not yet certain. The protection of critical infrastructure and the impact of AI are likely to feature highly in follow-up processes. The main controversies in remaining OEWG negotiations will be around modalities for the future cybersecurity process at the UN (i.e. institutional dialogue): should a future cybersecurity architecture deal mainly with implementing existing norms or negotiating new norms and legally binding instruments? Other open issues include the decision on topics and several thematic groups, and the inclusion of other actors through multistakeholder provisions (which will remain a central stumbling stone). One thing appears clear: the process will be continued, and the next one will likely be a permanent mechanism rather than time-limited. By July, the following outcomes are possible: Although time is not on the side of negotiators, there will be quite a few activities in the next six months, giving states more discussion opportunities. An informal town hall meeting to discuss the next mechanism will be held before the tenth substantive session scheduled for February. The OEWG’s schedule for the first quarter of 2025 includes the Global POC Directory simulation exercise, an example template for the Global POC Directory, and reports on the Global ICT Security Cooperation And Capacity-Building Portal, and the Voluntary Fund. Further, the chair can schedule additional intersessionals if deemed necessary. The race between cryptography and quantum computing In 2025, governments and companies will ramp up preparations for quantum computing, a technology poised to render current encryption obsolete. To address this, the US National Institute of Standards and Technology (NIST) introduced a post-quantum cryptography standard in 2024, featuring algorithms designed to withstand quantum attacks. This proactive policy approach demonstrates how society can tackle the uncertainties of emerging technologies like quantum computing. Meanwhile, the EU’s ‘Chat Control’ initiative dominates current encryption policy debates. It mandates tech platforms to scan content for illegal activities but faces strong opposition over privacy and human rights concerns. In December 2024, 10 member states rejected the proposal. Efforts to revive it through ‘Chat Control 2.0,’ which introduces ‘upload moderation’ requiring user consent for message scanning, are unlikely to succeed. Critics argue it fails to address the core issue of undermining encryption and creating security vulnerabilities. Major platforms like WhatsApp, Signal, Telegram, and Threema have threatened to exit the EU market if forced to weaken encryption protections. The implications of AI for international peace and security are typically tackled separately from the broader discussions on AI governance at the UN level. In December 2024, the UN General Assembly adopted the first-ever resolution on AI in the military domain, affirming the applicability of international law to systems enabled by AI in the military domain and encouraging member states to convene exchanges—at multilateral and multistakeholder levels—on responsible application of AI In the military domain. In 2025, the UN Secretary-General will have to follow up on this resolution, as he was requested to ‘seek the views of member states and observer States on the opportunities and challenges posed to international peace and security by the application of AI in the military domain, with specific focus on areas other than lethal autonomous weapons systems (LAWS)’; a substantive report summarising those views and cataloguing existing and emerging normative proposals will have to be submitted to the General Assembly at its 80th session, starting in September 2025. The GGE on emerging technologies in the areas of LAWS (convened yearly as a group since 2017) will continue its work in 2025, with two meetings planned for March and September. The group is tasked with ‘considering and formulating, by consensus, a set of elements of an instrument, without prejudging its nature, and other possible measures to address emerging technologies in the area of LAWS’. In 2024, the GGE worked on a so-called ‘rolling text’, which outlines provisional rough consensus on several formulations on issues such as the characterisation of a LAWS; applicability of international humanitarian law; human control and judgement as essential with regard to the use and effects of LAWS; several prohibitions on the use of LAWS, including a prohibition of employing LAWS that operate without context-appropriate human control and judgement; obligations for states prior to potential employment and as applicable throughout the entire life cycle of LAWS; and obligations for states to ensure human responsibility and accountability. With the increasing ‘digitalisation’ of ongoing conflicts, the applicability of international humanitarian law (IHL) in the cyber realm is set to become more prominent. A 2024 report by the International Committee of the Red Cross (ICRC) highlights two key challenges: In 2025, IHL and cyber conflicts are likely to be central to several cases before the International Court of Justice (ICJ), including South Africa vs. Israel, Nicaragua vs. Germany, and Ukraine vs. Russia.
Tech and a global decline of human rights The protection of tech-related human rights deteriorated in 2025, mirroring a broader global decline. The Trump administration deprioritised inclusive frameworks—including gender and other rights—while maintaining a singular focus on freedom of expression. Concurrently, major technology companies systematically dismantled content governance policies that had been built around human rights principles. The narrow focus of AI rights discourse While AI remained prominent in human rights discussions, the debate was largely confined to traditional narratives of bias and algorithmic ethics. However, the human rights community devoted less attention to AI’s profound impact on other fundamental areas—such as education, cognitive development, and social cohesion—that define the very fabric of communities. According to all indications, in 2025, Trump’s presidency will deprioritise human rights compared to the Biden administration’s focus. As tech companies retreat from content moderation, they will likely backpedal on the impact of their platforms on human rights. While EU countries will likely increase their focus on human rights in the digital age, addressing issues such as AI ethics, surveillance, and the impact of technology on privacy and freedom of expression. The EU will aim to ensure global relevance of EU’s regulations of digital relevance: GDPR, AI Act, DSA, etc. AI will be raised on the agenda of the UN Human Rights Council and other initiatives and organisations dealing with human rights. AI will bring new angles to reshaping ‘traditional’ human rights, such as freedom of expression and privacy protection. In addition, it will foster new types of human rights dilemmas. For instance, human identity will be highly relevant as individuals are impersonated through AI to mimic human looks and voices. Here, the dilemma is whether the question of identity can be covered by privacy protection or if it will require a new set of legal and policy rules. The rapid development of neurotechnologies, spurred by AI and biotech advancements, will bring neurorights to the forefront of human rights agendas. The 2024 report on neurotechnology and human rights of the UN Human Rights Council Advisory Committee, along with UNESCO’s work on the ethics of neurotechnology, will likely catalyse new international norms and regulations to protect cognitive liberty, mental privacy, and the integrity of the human mind. While AI will bring risks across human rights, there are areas where AI can help realise human rights. For example, AI can and should play a crucial role in increasing the well-being of people with disabilities. In the governance realm, the main focus should be on developing usability standards for people with disabilities. The main regulatory development on technology and disabilities will be the entry into force on 28 June 2025 of the European Accessibility Act, which will impose legal liability for businesses for providing equal access to digital products and services. This is a placeholder tab content. It is important to have the necessary information in the block, but at this stage, it is just a placeholder to help you visualise how the content is displayed. Feel free to edit this with your actual content. This is a placeholder tab content. It is important to have the necessary information in the block, but at this stage, it is just a placeholder to help you visualise how the content is displayed. Feel free to edit this with your actual content. The AI Bubble: Inflated, but not burst (yet) The AI bubble has expanded dramatically, fueled by both soaring expectations and immense investment. By the end of the year, a shared understanding emerged of the significant gap between the capital deployed and the relatively low revenue generated by AI. The critical question of how to manage this financial overhang has been decisively postponed to 2026. The securitisation of the tech economy The traditional division between digital technology and national security has continued to blur. “Digital economic security” is now central to policy agendas worldwide, leading to a growing web of sanctions and restrictive regimes governing cross-border technological exchange. Cryptocurrency’s counterintuitive year Bitcoin’s value trajectory in 2025 defied easy prediction. Despite the arrival of a crypto-friendly Trump administration and the introduction of favourable U.S. regulations—factors expected to drive a rapid price increase—the value of Bitcoin actually declined. It fell from $94,439.99 at the start of the year to $88,854.54 by January 31st, 2025, highlighting the complex and often counterintuitive dynamics of the cryptocurrency market. The 2025 prediction was prescient in identifying the key domains. In reflection, the year confirmed that technical standards are now inextricably linked to geopolitics, economic resilience, and ethical governance. They functioned not just as a “safety net” but as strategic assets and diplomatic tools. The grand narrative of 2025 was the struggle between forces using standardisation to maintain global interoperability and those using it to construct sovereign digital realms with controlled gateways. The outcome was not a single, unified standards regime, but a more complex, layered landscape where foundational protocols like TCP/IP remain universal, while higher-layer standards increasingly reflect the world’s political and economic divisions. AI standardisation: The push for open-source AI standards emerged as a particularly vibrant and contentious domain, aiming to prevent the locking down of foundational technology by a few corporate actors. Digital Public Infrastructure (DPI) standards: Standardisation became the linchpin for donor funding, vendor interoperability, and avoiding vendor lock-in. The ITU and World Bank were deeply involved in creating modular DPI standards frameworks, turning this into one of the most impactful standards stories of the year, directly affecting billions of citizens. Standardisation became the linchpin for donor funding, vendor interoperability, and avoiding vendor lock-in. The ITU and World Bank were deeply involved in creating modular DPI standards frameworks, turning this into one of the most impactful standards stories of the year, directly affecting billions of citizens. In 2025, social media moderation is undergoing a radical shift as major platforms enthusiastically adopt a “hands-off” approach, replacing centralised fact-checking with crowd-sourced systems like Community Notes. This move is primarily driven by a desire to reduce the substantial costs and liabilities associated with maintaining large trust and safety teams. In effect, platforms are transferring the burden of content judgement to users, betting that collective wisdom can ensure reliability. However, this experiment raises critical questions about its effectiveness against AI-generated misinformation, coordinated manipulation, and the overall decline in content quality, as platforms effectively abdicate their traditional gatekeeping role. In stark response to this perceived platform failure, Australia has implemented one of the world’s most restrictive policies: a ban on social media use for under-16s. This drastic measure reflects a complete loss of trust in platforms’ self-regulation and safety tools. The agreement of Meta and TikTok to comply underscores a pragmatic acceptance of state-imposed access restrictions, validating a regulatory model focused on user protection through prohibition rather than improving the online environment itself. This dynamic fuels the central geopolitical tension in digital governance: the clash between the EU and US-based tech giants. The EU, through fines such as the one levied against Platform X for breaches of the Digital Services Act, is aggressively enforcing a legalistic, rights-based framework that demands robust content moderation. This directly counteracts the platforms’ cost-saving “hands-off” strategy. The resulting conflict creates a significant regulatory battleground, pitting the EU’s stringent oversight against the US’s more libertarian tech ethos and the platforms’ commercial interests, leading to a fragmented and contentious global internet. The central policy process was the WSIS+20 review, which assessed two decades of digital progress with mixed results. While billions have been connected – raising all boats, though some faster than others – digital divides have persisted and widened, particularly with recent AI advancements. Alongside this, a major development was the Hamburg Declaration on Responsible AI for the SDGs, aiming to steer global ethical AI development through explicit principles of inclusivity and sustainability. AI Commons featured prominently during the February AI Summit in Paris (February 2025), focusing on open data in climate/health/education, and open-source AI platforms. Sectoral development activities include: An explosive energy demand of AI marked last year. With data centre electricity consumption projected to double to 1,000 TWh – equivalent to Japan’s total usage – the industry pursued radical innovation and energy sourcing. Concrete initiatives moved beyond pledges to tangible projects. Microsoft’s plan to restart a nuclear reactor at Three Mile Island and Google’s order of advanced reactors from Kairos Power exemplified the turn to next-generation nuclear to ensure clean, baseload power. Simultaneously, operational tactics like power capping and carbon-aware computing became standard practice, actively shifting workloads to optimise for renewable energy availability. In parallel, the circular economy transition gained regulatory teeth and technical sophistication. While global e-waste volumes remained alarming, binding frameworks emerged. The updated Basel Convention amendments, effective January 2025, established a stricter global regime for the movement of e-waste across national borders, despite implementation inconsistencies between OECD and non-OECD countries. Technologically, companies like Cisco set a benchmark by reusing or recycling nearly 100% of returned products, and AI itself was deployed to optimize recycling streams and identify reusable components, targeting waste reductions of 16-86%. The water intensity of AI also became a measurable crisis and a driver of innovation. The statistic that global AI demand could account for half of the UK’s annual water usage underscored the scale. In response, advanced cooling technologies, such as immersion cooling and liquid-to-liquid heat exchangers, saw accelerated adoption, offering up to 55% reductions in water usage. Regulatory pressure crystallized in mandates like Germany’s Energy Efficiency Act, requiring data centers to achieve 100% renewable energy by 2027, and in disclosure laws in California and the EU forcing transparency on water and energy use. At the international level, the anticipated advisory opinion from the International Court of Justice (ICJ) on state climate obligations promised to add significant legal and moral weight to the discourse, potentially influencing how environmental harms linked to digital infrastructure are framed as human rights issues.
Large Language Models became commoditised in 2025, with hundreds launched into the market. The significance of this shift was crystallised in the reaction to Apple’s cautious approach to AI investment, as highlighted in a November Bloomberg analysis:
With AI now a commodity, powered by open-source models, development is shifting toward smaller, specialized models tailored to individual companies, communities, and even countries. A growing number of organizations realize that the key to effective AI lies not in accessing the most cutting-edge LLM, but in leveraging the unique knowledge and data they already possess.
AI transformation: The commodity within, the cultural shift beyond
Geopolitics
Semiconductors
Geoeconomics
Geoemotions
Syncing GDC and WSIS
The Hanoi Cybercrime Convention
UN Cybersecurity partnership framework
Encryption
Military uses of AI and LAWS
International humanitarian law
