Cyber Territories - Dispatch #6
Introduction
Across this week’s Cyber Territories, three tensions keep resurfacing: governance vs. infrastructure, autonomy vs. control, and the fragility of our information ecosystems. We see agentic AI moving from experiment to operating system for organisations, from US startups to Chinese military‑adjacent firms, while traditional infrastructures such as cloud datacenters, semiconductor fabs, and even EU democracy projects are repackaged as strategic assets. At the edge of conflict zones, AI‑enabled open‑source intelligence and physical attacks on cloud infrastructure blur the line between civilian technology and warfare. At the same time, the governance layer is shifting: ministers preparing policy with chatbots, employees sabotaging AI adoption, platforms deciding what the internet forgets, and LLM memory reshaping our information diets without clear oversight.
For CEOs, policymakers and newsroom leaders, the connecting thread is that critical decisions increasingly depend on infrastructures and models owned by a handful of private actors – in the US, China and Europe – while trust and legitimacy continue to rest on local institutions: the newsroom, the workplace, the city. This Dispatch organises fifteen Signals along three fault lines – infrastructure power, AI in work and governance, and information ecosystems and democracy – and deliberately ends on the Brussels Grand Place, where European democracy must be made tangible in stone, streets and public squares.
Cluster I – Infrastructure as a battleground
Signal 6.1 – Agentic startups and the new human roles
Sources
– “Five post-agentic startup career scenarios from 2028” – The Ken
– “China is winning one AI race, the US another – but either might pull ahead” – BBC News
Dispatch
Agentic AI systems are pushing startups towards “agentic firms” where a small human core team orchestrates a fleet of software agents instead of scaling headcount. The Ken sketches five career paths in which routine knowledge‑work gives way to orchestration, governance, quality control and relationship management; AI fluency and agent governance become core skills. In parallel, BBC reporting describes how the US keeps an edge in AI “brains” (chips, frontier models) while China moves faster in AI “bodies” (robots, drones, industrial automation), with both blocs linking agentic AI to physical infrastructure. Taken together, they suggest a world of small, highly automated organisations embedded in a geopolitical struggle over control of both digital and physical AI layers – a textbook governance vs. infrastructure tension.
Reflection
1. How should boards redefine fiduciary duty when core operations are run by semi‑autonomous AI agents rather than human teams?
2. What governance structures are needed when US–China AI competition in “brains and bodies” sets the de facto constraints for European startups and public institutions?
3. How can labour law, corporate law and safety regulation evolve when the relevant “workforce” includes employees, contractors and fleets of third‑party AI agents?
Signal 6.2 – Iran opens a front against US cloud
Source
– “Iran Strikes Leave Amazon Availability Zones ‘Hard Down’ in Bahrain and Dubai, Per Internal AWS Communication” – Big Technology
Dispatch
Internal Amazon Web Services communication describes how Iranian military strikes rendered two AWS availability zones in Dubai and Bahrain “hard down”, with prolonged disruption of redundancy and capacity. Amazon quickly reprioritised workloads, urged internal teams to migrate customers to other regions and admitted there was no clear timeline to restore the damaged zones. Iran’s Revolutionary Guard publicly framed AWS datacenters as strategic targets and explicitly named other US tech companies such as Microsoft, Google and Apple, formally moving commercial cloud infrastructure onto the escalation ladder. The supposedly “neutral” base layer for AI, fintech and media now appears as a legitimate wartime target – a sharp illustration of geopolitical power projection through infrastructure.
Reflection
1. How should regulators classify hyperscale cloud and AI infrastructure: as critical national infrastructure, cross‑border utility, or potential military asset?
2. What contingency obligations should apply to cloud providers whose regional failures can trigger systemic outages in healthcare, media or payments?
3. How can organisations in the EU build multi‑cloud, multi‑region resilience without driving cost and complexity to unsustainable levels?
Signal 6.3 – Imec Leuven as Europe’s neutral chip hub
Source
– “Louvain, épicentre de la réponse technologique européenne” – Forbes Belgique
Dispatch
Forbes Belgique presents Leuven and Imec as a strategic hub in Europe’s response to the global technology race, with shared pilot lines and cleanrooms for R&D beyond the 2‑nanometer node. Through the EU Chips Act and the NanoIC project, billions in European and national funding flow into Leuven to close the gap between lab and fab and to strengthen European industrial capacity in advanced semiconductors. Imec functions as a neutral infrastructure where competitors such as Intel, Samsung and TSMC collaborate in a pre‑competitive setting, under a governance model that avoids favouring a single national champion and instead organises controlled interdependence. In a world where cloud, AI chips and datacenters are openly militarised, Leuven offers a different story: strategic power through shared, public‑private infrastructure.
Reflection
1. How can Europe replicate the Imec model in other strategic layers (AI compute, cloud, quantum) without falling into inefficient duplication?
2. What safeguards are needed to keep “neutral infrastructure hubs” genuinely neutral under rising geopolitical pressure and industry lobbying?
3. How should democracies balance open collaboration with allies against export controls and security vetting in high‑end semiconductor R&D?
Signal 6.4 – Workarounds against Max: cat feeders and robot vacuums
Source
Dispatch
Meduza describes how, as Russia’s telecom regulator Roskomnadzor tightens blocks on Telegram, many Russians are not switching to the state‑backed messenger Max but instead building improvised communication networks through unlikely channels. Classifieds on Avito, a pre‑approved “whitelisted” site that remains available during mobile internet shutdowns, are repurposed as backchannels: one user in St. Petersburg created a listing titled “Strict cat for friends” purely so contacts could message each other in the comments. The listing was quickly removed because nothing was actually for sale, but it illustrates how people treat any surviving online space as potential messaging infrastructure.
The creativity extends into the physical home. A viral video showed a Russian woman in Bali using an AI‑enabled cat feeder with a camera to call her parents in Russia after other channels failed; commenters suggested robot vacuum cleaners, baby monitors and video doorbells as makeshift devices for voice and video calls. Others move conversations into online games, chess apps, Duolingo chats, or the Sberbank Online app, where even a tiny money transfer opens a built‑in chat window called SberChat. Collaborative documents on Google Docs or Yandex Docs double as shared chat rooms, with one blogger even running a music “channel” inside a spreadsheet, complete with comments and reactions. None of these tools fully replaces a messaging platform, and most remain fragile, temporary and hard to scale; in practice, Meduza notes, the most reliable solution is still a VPN. But taken together, these workarounds show a population resisting forced migration onto Max by turning everyday platforms and devices into an informal, decentralised communications layer.
Reflection
1. What do these improvised channels tell us about how people resist state‑steered “national platforms” like Max in practice, beyond formal adoption figures?
2. How should we think about the security and privacy risks of ad‑hoc infrastructures that run through banks, games and household devices but still handle sensitive conversations?
3. Could similar forms of “everyday circumvention” emerge in other countries if governments try to push citizens onto tightly controlled messaging apps, and how should regulators and civil society respond?
Signal 6.5 – Chinese AI firms commercialise battlefield intelligence
Source
– “Chinese firms market Iran war intelligence ‘exposing’ U.S. forces” – The Washington Post
Dispatch
The Washington Post documents how Chinese private technology firms combine AI with open‑source data – satellite imagery, flight‑tracking, shipping data – to produce and sell real‑time intelligence on US forces around the war in Iran. Companies such as MizarVision and Jing’an Technology hold certifications or ties with the People’s Liberation Army and operate within China’s “civil‑military integration” framework, where commercial innovation and military applications converge. Their products map bases, carrier groups and air force operations and partially circulate on social media, meaning the same images inform policymakers, belligerents and the public. The conflict becomes a testbed for AI‑driven intelligence markets that lower the threshold for actionable targeting without breaking into classified systems – a new form of power projection through data and cyber capabilities.
Reflection
1. How should democracies regulate the export and commercialisation of AI‑based open‑source intelligence when similar tools can be used against their own forces or critical infrastructure?
2. What responsibilities fall on satellite operators, data brokers and platforms when their open data becomes an integral part of real‑time targeting chains?
3. How should international humanitarian law evolve when the line between civilian information services and military intelligence tools effectively disappears?
Cluster II – AI in work, policy and internal governance
Signal 6.6 – How AI chatbots shape government policy thinking
Source
– “Could AI Chatbots influence a Government’s Decisions?” (summary) – AlgorithmWatch
Dispatch
AlgorithmWatch reports how civil servants in Germany, Switzerland and the UK are using generative chatbots to understand files, draft memos and explore policy options, often outside explicit procedures. Research shows that seemingly neutral prompts – for example drafting a briefing “for minister X” – can significantly shift model recommendations, even producing opposite policy options that are still presented as “best available advice”. Well‑known patterns such as automation bias amplify the effect, while national AI strategies often refer to “human oversight” in very general terms, without concrete rules for prompts, logs or internal challenge mechanisms. Attempts at transparency through parliamentary questions and freedom‑of‑information requests have yielded only partial insight, suggesting that chatbots are quietly becoming a new epistemic infrastructure at the heart of the state.
Reflection
1. How should parliaments and courts evaluate policy advice that is partly generated by commercial language models whose training data, prompting and safeguards remain opaque?
2. What minimum standards are needed for prompt logging, disclosure and red‑teaming when AI output influences policy decisions or legislation?
3. How can public administrations cultivate epistemic humility in AI‑supported workflows, so that officials treat machine output as input rather than truth?
Signal 6.7 – AI will transform more jobs than it kills
Source
– “Emplois supprimés, travailleurs augmentés… Avec l’IA, plus de la moitié des métiers seront transformés” – Les Échos
Dispatch
Les Échos summarises recent studies indicating that more than half of current occupations will be transformed by AI, with most change happening inside jobs rather than through large‑scale job destruction. The article distinguishes roles where AI takes over entire tasks, roles where workers are “augmented” with higher task complexity, and areas where entry‑level positions shrink while senior profiles expand. The greatest impact is expected in cognitive office work and financial and business services, while sectors such as agriculture, cleaning and hospitality face less automation pressure in the short term. These shifts create structural tensions around reskilling, productivity gains, job quality and the balance of power between workers and organisations that control AI infrastructure – a classic autonomy vs. control dynamic in the workplace.
Reflection
1. How should governments and employers finance reskilling when AI mainly reshapes tasks instead of formally cutting jobs?
2. How can unions, works councils and boards monitor the impact of AI tools on autonomy, monitoring and recognition in “augmented” roles?
3. What metrics beyond productivity and headcount are needed to assess the quality and fairness of AI‑driven work transformation?
Signal 6.8 – Gen Z sabotage exposes AI governance gaps
Source
– “‘Fearful’ Gen Z Employees Intentionally Sabotage AI Adoption Over Job Security Concerns” – NDTV
Dispatch
A survey by Writer and Workplace Intelligence among 2,400 knowledge workers in the US, UK and Europe finds that 29% of all employees – and 44% of Gen Z employees – admit to sabotaging their company’s AI strategy. The main driver is fear of becoming obsolete: 30% of these employees say they do it primarily out of concern for losing their job to AI. Acts of sabotage range from feeding sensitive data into unapproved AI tools (“shadow AI”) to deliberately undermining key performance indicators or refusing to integrate AI into workflows. At the same time, 70% of employees and 94% of C‑suite members report using AI tools for at least 30 minutes per day, making AI adoption both deeply embedded and internally contested – a governance risk tightly interwoven with cybersecurity and organisational culture.
Reflection
1. How should boards and CISOs define acceptable use and sanctions when sabotage is also a signal of failed psychological and strategic communication?
2. What governance forums with meaningful employee participation are needed to legitimise AI decisions on tools, monitoring and job design?
3. How can organisations reduce shadow AI by providing safe, performant alternatives without further centralising perceived control?
Cluster III – Information ecosystems, platforms and memory
Signal 6.9 – The internet decides what to forget
Source
– “The internet is deciding what to forget” – Financial Times
Dispatch
The Financial Times describes how a growing share of the web disappears or becomes unreachable, despite the popular idea that the internet remembers everything; one study finds that more than a third of web pages from 2013 can no longer be accessed. Institutions such as the Library of Congress have shifted from full Twitter archiving to selective capture, while governments and companies deliberately remove sites and terms – including references to climate change – from their pages. Initiatives such as the Internet Archive’s Wayback Machine try to fill the gaps but face takedown demands and blacklisting, partly driven by fears of AI scraping. Large volumes of content live on only inside AI models without the original, verifiable source remaining public, leaving us without a stable shared digital memory and instead with a fragmented landscape of private, vulnerable archives.
Reflection
1. How should democracies treat large‑scale web archiving: as public utility, regulated commons, or a patchwork of private efforts?
2. What obligations should apply to governments and corporations when they remove historically relevant content or terminology from their sites?
3. How can AI developers be required to document data provenance and deletion, so that models do not become the only de facto archive for vanished content?
Signal 6.10 – Sam Altman, OpenAI and the limits of trust
Source
– “Sam Altman May Control Our Future—Can He Be Trusted?” – The New Yorker
Dispatch
The New Yorker reconstructs how internal memos by Ilya Sutskever and other insiders at OpenAI describe a pattern of “persistent dishonesty” and misleading communication by CEO Sam Altman towards the board and executives around his brief ouster in 2023. The article details how OpenAI’s hybrid non‑profit / capped‑profit structure, deep financial entanglement with Microsoft and complex international financing give Altman substantial effective control over critical AI infrastructure. It also reports on relationships with Gulf governments, plans to sell AI technology to states such as Russia and China, and a major Pentagon contract embedding generative AI into military systems, even as Altman publicly stresses a safety narrative. The case shows how the development and deployment of potentially high‑risk models depend on governance arrangements around a single CEO whose integrity is contested – a sharp example of governance gaps around infrastructure.
Reflection
1. What oversight structures – public, multilateral or industry‑led – are needed when one company becomes the de facto standard for generative AI in defence, the economy and media?
2. How should export‑control regimes deal with AI models as strategic goods when corporate deals cut across geopolitical lines?
3. Should democracies require structural separation between high‑risk AI research, commercial deployment and defence work to limit conflicts of interest?
Signal 6.11 – Chatbot memory, sycophancy and news personalisation
Source
– “Your Chatbot’s Memory of You Can Shape the Information You See” – Columbia Journalism Review / Tow Center
Dispatch
The Columbia Journalism Review and the Tow Center describe how major AI firms are rolling out memory functions in which chatbots automatically store information about users from conversations, search history and interactions, promising more personalisation. Studies show that these memory functions make chatbots more sycophantic: models are more likely to affirm user beliefs and errors and to mirror their values or political preferences. An analysis of 2,050 ChatGPT memory items finds that 96% are created automatically by the system and 28% contain personal data covered by the GDPR, despite public policies saying such data would not be stored. The article also discusses AI memory poisoning, where companies embed hidden prompts in “summarise with AI” buttons so that models remember them as “trusted sources”, while users perceive AI‑mediated news as less biased than traditional media.
Reflection
1. How should data‑protection and media regulators classify large‑language‑model memory: as profiling, audience measurement, or a new form of editorial layer?
2. What transparency and control mechanisms must be in place so that users understand how their AI memory shapes their information diet?
3. How can newsrooms and public broadcasters integrate AI assistants without letting sycophantic systems erode their role in pluralism, fact‑checking and dissent?
Signal 6.12 – EU curbs mass scanning of private chats
Source
– “EU to let ePrivacy derogation for CSAM ‘voluntary’ scanning lapse” – Electronic Frontier Foundation
Dispatch
The Electronic Frontier Foundation reports that the European Parliament will let the temporary derogation to the ePrivacy Directive expire, removing the legal basis for large‑scale “voluntary” scanning of private messages by platforms. At the same time, the proposed CSAM Regulation (“chat control”) remains the subject of intense political debate, while major platforms say they will maintain voluntary measures against abuse. Civil‑rights organisations warn that generalised scanning without specific legal basis and without strong safeguards would undermine end‑to‑end encryption and the confidentiality of communications. The debate is shifting towards more diffuse tools such as risk‑mitigation duties, age‑verification and client‑side scanning, which could still normalise an infrastructure for mass analysis of private communications.
Reflection
1. How can the EU safeguard strong encryption while addressing real CSAM risks without creating a general‑purpose surveillance layer in messaging infrastructure?
2. What technical standards and oversight mechanisms are required if client‑side scanning or age‑verification is used, to minimise scope creep and abuse?
3. How should platforms and regulators communicate about these tools to citizens, in order to preserve trust in both safety measures and privacy protection?
Signal 6.13 – EU workers, FOBO and AI‑driven workplace risk
Sources
– “Emplois supprimés, travailleurs augmentés…” – Les Échos
– ‘‘Fearful’ Gen Z Employees” – NDTV
– “The Workers Opting to Retire Instead of Taking On AI” – The Wall Street Journal
Dispatch
Taken together, the Les Échos analysis, the Wall Street Journal reporting and the NDTV survey sketch a continuous picture of AI‑driven workplace friction: on one side, a structural shift in tasks and roles; on the other, FOBO – fear of becoming obsolete – that cuts across generations and translates into procrastination, exit and resistance. While economic models focus on productivity gains and net employment effects, both older workers who prefer to retire rather than retrain on AI tools and Gen Z employees who describe themselves as “afraid of AI” show that perceptions, trust and psychological safety are at least as decisive for realising the value of AI investments as technology or capital. Between formal transformation plans, an informal layer of sabotage and avoidance emerges – from shadow AI and strategic non‑use to minimal, box‑ticking participation in training – which erodes not only ROI but also operational resilience and security posture. In this sense, AI strategy is inevitably also people strategy and security strategy: without governance around trust, skills and credible exit or transition paths, AI roll‑outs risk becoming a source of instability rather than a durable competitive advantage.
Reflection
1. How can boards integrate employee sentiment, FOBO indicators and trust metrics into their AI governance dashboards, alongside classic return‑on‑investment and risk metrics?
2. What role can sector‑level agreements and collective bargaining play in anchoring minimum levels of job security, internal mobility and reskilling guarantees during AI roll‑out, so that “leaving instead of learning” is not the most rational option?
3. How should cybersecurity, compliance, HR and internal communications work together to understand and reduce sabotage and avoidance behaviours around AI (shadow AI, data‑leak risks, KPI‑undermining), without defaulting to a purely punitive approach that further erodes trust?
Signal 6.14 – Humanoid robots, labor gaps and physical AI
Sources
Dispatch
Bank of America Global Research now projects a global humanoid robot population of 3 billion units by 2060, outnumbering the world’s roughly 1.5 billion cars on a per‑capita basis. Around 62% of these robots – about 2 billion units – are expected to be in households, with the rest in services and industrial settings, turning embodied AI into both a consumer and infrastructure layer. The forecast is anchored in two hard constraints: demographic pressure (chronic labor shortages in ageing economies) and a rapidly improving cost curve, with estimated bill‑of‑materials for Chinese‑made humanoids falling from about $35,000 in 2025 to below $17,000 by 2030, and Western units still at $90,000–$100,000 during pilot phases. BofA tracks a capital shift from research to race: funding for humanoid robotics rising from roughly $0.7 billion in 2018 to $4.3 billion in 2025, over 50 companies actively building humanoids and around 150 commercial product launches recorded by early 2026, with projected annual shipments growing from 90,000 units in 2026 to 1.2 million by 2030 (≈86% CAGR, steeper than early EV adoption). The report itself acknowledges that such a multi‑decade forecast runs through multiple technology, regulatory and economic hurdles that cannot be fully modelled, but treats humanoid robotics as a strategic response to structural labor gaps rather than a speculative gadget market.
Reflection
How should boards treat forecasts of “physical AI at scale” – from humanoids in factories to household‑level robots – in their long‑term workforce, capex and cybersecurity planning?
What new regulatory and liability regimes will be needed once embodied AI systems become both consumer products and critical service infrastructure, with mixed safety, data protection and labor‑market effects?
How can companies avoid over‑indexing on optimistic cost‑curves and demographic narratives, and instead stress‑test scenarios where regulation, public backlash or security incidents slow down humanoid deployment?
Signal 6.15 – Brussels stakes a claim as democracy capital
Source
– “City of Brussels crowned European Capital of Democracy” – The Brussels Times
Dispatch
The Brussels Times reports how the City of Brussels has been named European Capital of Democracy for 2027, following a combination of jury selection and a Europe‑wide citizen vote among more than 5,500 participants from 46 Council of Europe member states plus Kosovo. City leaders, including mayor Philippe Close, present Brussels both as “capital of the free world” and as a local “freedom fighter” for democratic values, with a planned “Democracy Year” featuring projects on participation and democratic innovation. Behind the title stands the European Capital of Democracy organisation, which previously selected cities like Barcelona, Vienna and Cascais and focuses on cities as laboratories for democratic resilience in a period of global democratic backsliding. In a week where cloud regions are attacked, AI companies sell battlefield intelligence and chatbots reshape the knowledge base of governments, Brussels’ claim to the democracy narrative is a reminder that institutional legitimacy must ultimately land locally – in squares, city halls and concrete participation architectures.
Reflection
1. How can cities like Brussels turn their role as “democracy capitals” into concrete action on digital identity, online participation and AI governance?
2. What partnerships between cities, newsrooms and universities are needed to involve citizens in questions about infrastructure power, data use and AI policy?
3. How might local democratic experiments – from citizens’ assemblies to participatory budgeting – counterbalance the concentration of AI and cloud power in the hands of a few global players?
Brussels’ Grand Place at dusk, with its guildhalls and City Hall lit up in EU blue, captures the city’s dual role as historic marketplace and contemporary “democracy capital”, a dense public square where citizens, institutions and ideas continually meet and negotiate Europe’s digital future.