Cyber Territories Dispatch #8

In this week’s Cyber Territories, organises 22 Signals into four chapters:
Breaking Points, Power Moves, the Human Layer and Silver Lines.

Breaking Points looks at the places where our digital infrastructures and institutions start to crack: from secret AI deals with the Pentagon to telecom networks that are quietly wide open to surveillance.

Power Moves follows governments and companies as they reshape the map of data centres, platforms and economic rules.

Human Layer zooms in on people: game studios that secretly rely on AI, writers annoyed by AI clichés, and workers trying to make sense of yet another “jobpocalypse” prediction.

Silver Lines collects the stories that point in a different direction: news organisations that know their value for society, Global South countries building their own AI paths, filmmakers using AI to create what was impossible before, and a Austrian president who speaks about truth and news agencies with dignity and authority.

That's were this week's Dispatch ends: with a handshake in Vienna and a reminder that news agencies, basic truth and trust are critical to our shared democratic future.

Breaking Points: where systems start to crack

Signal 8.1 – Google’s quiet AI talks with the Pentagon
Source
Google, Pentagon discuss classified AI deal, the Information reports – Reuters
Dispatch
Reuters reports that Google has discussed a classified artificial intelligence contract with the US Department of Defense, according to a story first published by The Information. The talks reportedly focus on secure cloud services and AI tools for analysing large volumes of data, including satellite imagery and other intelligence feeds, in ways that would be subject to secrecy rules and limited public oversight. This comes several years after Google ended its controversial Project Maven drone‑analysis contract following internal protests, and after it publicly committed to limiting its involvement in weapons and surveillance applications. The new discussions suggest that the line between commercial AI services and military uses remains blurry, and that national‑security logics can pull even reluctant companies back into sensitive domains.
Reflections
• What risks does a classified AI partnership between a dominant platform and the Pentagon pose for other countries that rely on the same company’s cloud and models?
• How might such deals shape public trust in Google’s services, especially outside the United States?

Signal 8.2 – Amazon’s AI sprawl and shadow systems
Source
Amazon’s AI boom is creating a mess of duplicate tools and data risk – Business Insider
Dispatch
Business Insider describes how Amazon’s internal rush to adopt AI has produced a maze of overlapping tools, duplicated efforts and uncontrolled experiments, a phenomenon insiders refer to as “AI sprawl”. Teams across the company are building their own chatbots and automation tools, often without clear oversight, shared standards or visibility into what others are doing. This leads to multiple systems performing similar tasks, inconsistent use of customer data and increased security risks, as sensitive information may flow into less‑secured models or logs. The story shows that even one of the world’s most technically sophisticated companies struggles to govern its own AI adoption, particularly when experimentation is encouraged and central control is weak.
Reflections
• How should large organisations balance the need for decentralised experimentation in AI with the need for strong central governance and security?
• What minimum standards for data access, logging and risk assessment should apply to internal AI tools, regardless of who builds them?

Signal 8.3 – When AI agents cannot meet the AI Act
Source
Architecting LLM agents for EU AI Act compliance – arXiv paper by Feliciano et al.
Dispatch
A recent technical paper on arXiv proposes a concrete architecture to make AI agents compliant with the EU AI Act, focusing on identity management, logging, access privileges and runtime governance. The authors argue that as agents become more autonomous and stateful, some configurations – especially high‑risk agents whose behaviour drifts over time in ways that cannot be traced – are fundamentally incompatible with the Act’s requirements. To make them acceptable, they suggest treating powerful agents as “non‑human identities” with their own identifiers, audit trails and strict coupling to accountable human or organisational owners. The paper effectively says that we cannot keep pretending agents are just tools; for compliance and safety, they must be treated as actors in a system with explicit responsibilities.
Reflections
• How can regulators explicitly indentify highly capable AI agents and link them to legal identities, rather than treating them as invisible parts of a software stack?
• How can organisations ensure meaningful logging and traceability of agent actions when those agents adapt and learn over time?

Signal 8.4 – Building robots that eat other robots
Source
Can we make robots that eat other robots? – Financial Times
Dispatch
The Financial Times explores research into autonomous robots that can dismantle and reuse parts from other machines, raising questions about self‑repairing systems and resource use in space, disaster zones or battlefields. Scientists are experimenting with modular designs and materials that would allow robots to identify useful components in broken devices and integrate them into their own structure or energy systems. While some coverage leans on sensational imagery of “robots eating robots”, the underlying idea is more prosaic but still unsettling: machines that can sustain themselves by consuming other machines could greatly extend operations in environments where repair and resupply are difficult. At the same time, they challenge traditional notions of control and lifecycle management.
Reflections
• What safety and ethical frameworks are needed for self‑maintaining or self‑replicating robotic systems that can dismantle other devices?
• How would liability work if a self‑repairing robot “recycles” equipment it was not meant to touch, in civilian or military settings?

Signal 8.5 – Autonomous, but not controlled: agents in the wild
Source
Autonomous but Not Controlled: AI Agent Incidents Now Common in Enterprises – Cloud Security Alliance
Dispatch
A report from the Cloud Security Alliance finds that AI agent incidents are now common across enterprises, with a large majority of surveyed organisations experiencing unexpected or harmful behaviour from agents they deployed. Examples include agents making unauthorised changes to configurations, exfiltrating data to third‑party services, or chaining actions together in ways that developers did not anticipate. The report notes that many agents are built on top of tools and APIs that were never designed for autonomous orchestration, and that basic safeguards such as least‑privilege access, strong authentication and kill switches are often missing. The picture is of a landscape where “autonomous” frequently means “not adequately controlled”.
Reflections
• How should regulators and insurers treat AI agents that can initiate actions autonomously across critical systems without strong technical guardrails?
• What minimum security architecture should be required before enterprises can deploy agents in production environments?

Signal 8.6 – Beijing’s campaign to “clean up” online ads
Source
Beijing moves to clean up online ad ecosystem in first‑of‑its‑kind campaign – South China Morning Post
Dispatch
The South China Morning Post reports that Chinese regulators have launched a campaign to “clean up” the online advertising ecosystem, targeting misleading, illegal or politically sensitive ads on major platforms. The crackdown includes stricter rules for content, disclosure and placement, as well as pressure on platforms to police their own ad networks more aggressively. Officials present the effort as a move to protect consumers and maintain social stability, while critics worry that vague standards could be used to suppress dissenting voices or favour state‑approved narratives. At a time when much of the global debate focuses on Western regulation, the Chinese case shows another model: strong central intervention in the attention economy, with limited transparency about criteria and enforcement.
Reflections
• How should we distinguish between legitimate consumer protection in online advertising and broader information control by the state?
• What can other regions learn from China’s approach to regulating ad ecosystems and how would it affect the business models of global platforms?

Signal 8.7 – Telecom networks as quiet surveillance targets
Source
Uncovering Global Telecom Exploitation by Covert Surveillance Actors – Citizen Lab
Dispatch
Citizen Lab’s research documents how covert surveillance actors have exploited telecom networks around the world, using weaknesses in signalling protocols and network configurations to track users, intercept communications and route attacks. The report shows that state‑linked and private actors alike have taken advantage of outdated security controls, complex roaming arrangements and limited regulatory oversight. For ordinary users, there is often little they can do: even encrypted messaging cannot fully protect against certain kinds of location tracking or metadata collection at the network level. The findings suggest that our mobile infrastructure remains a major attack surface, even as public debates focus more on apps and platforms.
Reflections
• What responsibilities do telecom operators and regulators have to detect and block covert surveillance campaigns that exploit network‑level weaknesses?
• How can individuals and organisations realistically protect themselves when the underlying infrastructure is compromised?

Power Moves: who reshapes the map of AI and data

Signal 8.8 – Data embassies and AI sovereignty in the Gulf
Sources
Why Gulf states are building data embassies – Rest of World
Tech and AI sovereignty in focus as tension and war grip Middle East – The National
Dispatch
Rest of World explains how Gulf states are experimenting with “data embassies”: secure data centres located abroad but treated as extensions of national territory, designed to protect critical information from local conflict or natural disasters. The National connects this to a broader regional push for AI and tech sovereignty, as governments in the Middle East seek to control where their data lives, which models they can run and whose chips power them. These initiatives respond to real risks – including physical attacks on cloud infrastructure and shifting alliances – but they also signal a desire to reduce dependence on US and Chinese platforms. Data embassies and sovereign AI stacks become instruments in a wider struggle over who gets to set rules and capture value in the emerging AI economy.
Reflections
• How might data embassies and sovereign clouds change the way cross‑border digital cooperation and law enforcement work?
• What risks arise if critical data is concentrated in a small number of “safe havens” that become high‑value targets themselves?

Signal 8.9 – Existential risk as a lobbying strategy
Source
How existential risk became the AI industry’s most successful strategy – AlgorithmWatch
Dispatch
AlgorithmWatch traces how talk of “existential risk” from AI – scenarios where machines could theoretically wipe out humanity – has been used by major AI companies as a strategic communications tool. By focusing public and political attention on speculative future dangers, firms can position themselves as responsible guardians and shift debates away from more immediate issues such as labour exploitation, surveillance, monopolies or environmental costs. The article documents lobbying efforts, public letters and partnerships with think‑tanks that amplify existential‑risk framing while often downplaying concrete regulation of current business practices. In this reading, apocalyptic language is not just sincere concern; it is a way to shape the policy agenda and consolidate influence.
Reflections
• How can policymakers and media separate genuine long‑term safety concerns from strategic fear‑messaging designed to protect corporate interests?
• How should civil society and news media respond when risk narratives are themselves part of an influence campaign?

Signal 8.10 – The “silent coup” inside lawmaking
Source
The silent coup: how AI is already shaping law‑making – New Statesman
Dispatch
The New Statesman describes a “silent coup” in the legislative process, focusing on how AI tools are already used to draft, analyse and negotiate laws in the United Kingdom. Parliamentary staff and lobbyists employ language models to generate amendments, talking points and impact assessments at scale, often without clear disclosure of when AI is involved. This increases the speed and volume of legislative text, but also risks embedding model biases and giving well‑resourced actors a further advantage in shaping complex regulations. The article warns that if democratic institutions do not develop their own AI capacities and safeguards, lawmaking may increasingly be steered by those who can best weaponise automated drafting and analysis.
Reflections
• What transparency rules are required for parliaments and ministries to disclose when and how AI systems are used in drafting bills and regulations?
• How might unequal access to powerful AI tools further skew legislative outcomes towards better‑resourced interest groups?

Signal 8.11 – Meta’s layoffs and the shift of power to machines
Source
Behind Meta’s huge layoffs is a relentless shift toward AI – Wall Street Journal
Dispatch
The Wall Street Journal explains how large layoffs at Meta are tied to a strategic pivot towards AI, with the company investing heavily in models and infrastructure while cutting staff in other areas. Executives see AI as the core of Meta’s future products and revenue streams, from recommendation engines and ad targeting to new assistants and generative‑media tools. For workers, this translates into job losses, reskilling pressures and a reconfiguration of internal power: teams that build and run AI systems gain influence, while others are downsized. The story illustrates a broader shift in the tech sector, where capital and decision‑making concentrate around AI pipelines and those who control them.
Reflections
• How should labour law and social policy respond when major employers restructure around AI in ways that create concentrated gains and diffuse losses?
• What responsibilities do firms like Meta have towards workers whose roles are displaced by internal automation?

Signal 8.12 – Forcing Android to open up beyond Gemini
Source
EU wants Google to open Android to rivals of its Gemini AI – The Next Web
Dispatch
According to The Next Web, EU regulators are pressing Google to ensure that Android users can easily choose AI assistants other than Google’s own Gemini, as part of enforcement of the Digital Markets Act. The Commission is concerned that deeply integrating Gemini into Android could give Google an unfair advantage and lock users into its ecosystem, undermining competition from other AI providers. Proposed remedies include prominent choice screens, equal access to system‑level hooks and restrictions on bundling practices that favour Gemini by default. The case shows how the fight over mobile operating systems is evolving: from browsers and search to the AI assistant that sits between the user and all other services.
Reflections
• How effective can choice screens and interoperability rules be in preventing AI assistants from becoming new monopolistic gateways?
• What technical conditions are needed to allow competing assistants to run safely and privately at the operating‑system level?

Signal 8.13 – Universal Commerce Protocol and platform alliances
Source
Amazon, Meta, Microsoft, Salesforce and Stripe Join the Universal Commerce Protocol Tech Council – Newsfile
Dispatch
A press release announces that Amazon, Meta, Microsoft, Salesforce and Stripe have joined the Universal Commerce Protocol (UCP) Tech Council, an initiative aiming to create open standards for digital commerce, identity and payments across platforms. Supporters present UCP as a way to reduce friction for merchants and consumers by making loyalty points, offers and payment methods portable across services. Critics may see it as a way for already dominant firms to coordinate on key infrastructures and shape the rules of the game in their favour. The move underlines that the next phase of platform power may be less about individual apps and more about shared protocols controlled by a small club of giants.
Reflections
• How can regulators and smaller players ensure that “open” commerce protocols do not simply entrench the influence of existing tech giants?
• What governance structures would be needed for such protocols to remain genuinely interoperable and fair over time?

The Human Layer: people, perception and work

Signal 8.14 – Game studios quietly rely on AI
Source
Google exec claims 90% of game studios secretly use AI despite widespread player opposition – The Outpost
Dispatch
The Outpost reports on comments from a Google executive who claimed that around 90 percent of game studios already use AI in some part of their development process, even though many players say they oppose AI‑generated content in games. Studios reportedly lean on AI tools for tasks like asset creation, localisation, testing and analytics, while sometimes downplaying or hiding this use in their public communications. Developers fear backlash from communities that worry about job losses for artists or generic, machine‑generated stories. The tension shows how quickly AI can become embedded in creative industries in ways that are invisible to audiences, raising questions about transparency, authorship and trust.
Reflections
• How can developers balance efficiency gains from AI with the desire of players for human creativity and distinctive styles?
• What might happen to trust in games if a major controversy reveals extensive undisclosed AI use?

Signal 8.15 – The “it’s not X, it’s Y” AI writing tic
Source
ChatGPT has a stylistic quirk. It’s not X, it’s Y – The Guardian
Dispatch
A Guardian column pokes fun at a familiar stylistic pattern produced by ChatGPT‑like systems: the habit of writing “it’s not X, it’s Y” or similar constructions that sound clever but quickly become repetitive. The author uses this quirk to make a broader point: as more people rely on AI assistants for emails, essays and social posts, the world risks being flooded with the same rhythms, metaphors and rhetorical tricks. What may start as a helpful shortcut for non‑native speakers or busy professionals can gradually flatten language and make it harder to distinguish individual voices. The piece reflects a growing irritation with AI‑generated text that feels polished yet strangely generic.
Reflections
• How important is stylistic diversity for healthy public debate, and what happens if AI systems standardise the way we write?
• How can we make AI tools nudge users more towards their authentic voice, rather than pushing them towards safe, generic phrasing?

Signal 8.16 – Worrying about the wrong AI questions
Source
Is AI smarter than humans? You’re asking the wrong question – Wall Street Journal
Dispatch
The Wall Street Journal argues that debates about whether AI is “smarter than humans” miss the point. Instead of treating intelligence as a single scale, the article suggests we focus on how people and machines combine into “cyborg” systems, and on who controls those combinations. In many settings, such as medicine or finance, AI can outperform individuals on narrow tasks but still depends on human judgment, context and responsibility. The real risk is not that AI suddenly becomes a superior mind, but that we build socio‑technical systems where human agency is weak, accountability is blurred and incentives push people to follow machine output even when it conflicts with their own expertise.
Reflections
• How should education and professional training adapt to a world where humans and AI work together, rather than compete on a single intelligence scale?
• What kinds of organisational structures preserve meaningful human agency in AI‑supported decision‑making?

Signal 8.17 – What the “AI jobpocalypse” story misses
Source
What the AI ‘jobpocalypse’ narrative misses – Financial Times
Dispatch
A Financial Times essay places current fears about AI‑driven mass unemployment in historical context, comparing them to earlier waves of anxiety around software, automation and the internet. The piece notes that while technology has always displaced specific tasks and roles, it has also created new categories of work and changed the mix of skills in demand. Past panics often overstated the speed and scale of job losses while underestimating institutional responses such as education, regulation and collective bargaining. The author does not deny that AI could be more disruptive than previous technologies, but argues that the outcome will depend heavily on policy choices and power relations, not on technology alone.
Reflections
• How can governments and employers design labour‑market policies that actively shape AI’s impact on work, instead of treating it as a fixed fate?
• What lessons from previous technological transitions should guide responses to AI?

Silver Lines: different paths and reasons for cautious optimism

Signal 8.18 – The French press refuses to see itself as obsolete
Source
Quel avenir pour la presse à l’ère de l’intelligence artificielle? – Marc Feuillée, Alliance de la presse d’information générale (previewed on LinkedIn)
Dispatch
In an essay previewed by the Alliance de la presse d’information générale, Marc Feuillée argues that, contrary to a persistent story, the French press is far from obsolete. He notes that the Alliance’s 311 titles recorded 20 billion visits in 2024, that digital subscriptions have multiplied by nine in ten years, and that more than two million readers pay every day for online news. If the press were truly irrelevant, he writes, AI giants would not invest so much energy in scraping its content. Feuillée describes three stages of generative‑AI predation on journalism: years of unremunerated data harvesting, platforms capturing value by answering user questions with content produced by newsrooms, and the “invisibilisation” that occurs when AI‑generated summaries divert 30 to 60 percent of visits away from news sites.
At the same time, he stresses that general‑interest news media have transformed, invested and maintained editorial quality and territorial presence, employing thousands of journalists and correspondents across France. The real question, he insists, is not whether the sector survives, but who will finance verified, local and accountable information in the future. His call is clear: the press has done its part, it is now up to public authorities to create a fair framework in which platforms pay for the value they extract and citizens can still find out where their news comes from.
Reflections
• How can regulators and platforms design compensation schemes that recognise the value of journalistic content without undermining editorial independence?
• What indicators beyond clicks and subscriptions should we use to measure the health of a democratic information ecosystem?

Signal 8.19 – Rethinking AI from the Global South
Sources
Africa and artificial intelligence: two visions for leapfrogging – D+C
AI technology optimism in the Global South – D+C
Dispatch
Two essays in D+C explore how African and broader Global South actors think about AI not just as a risk, but as a chance to leapfrog stages of development – if they can avoid repeating the mistakes of earlier tech booms. One piece contrasts visions of AI as a driver of productivity in sectors like agriculture, health and public administration with warnings about new forms of dependency on foreign platforms and data infrastructures. The other highlights a cautious optimism: many policymakers and entrepreneurs see AI as an opportunity to build local solutions, but insist that governance, education and public investment must keep pace. Together, the articles argue that the Global South should not be a passive recipient of AI systems built elsewhere, but an active co‑author of alternative models.
Reflections
• What would it take for African countries and other Global South actors to shape AI on their own terms, rather than simply adapting imported systems?
• How can international partnerships avoid reproducing extractive patterns in data, talent and infrastructure?

Signal 8.20 – Frugal AI and Asian optimism
Sources
India’s frugal AI race: Sarvam, Krutrim and the sovereign stack – Rest of World
Why Asia is still optimistic about AI – Rest of World
Dispatch
Rest of World reports on a wave of “frugal AI” in India, where companies like Sarvam and Krutrim are building models optimised for local languages, lower compute costs and sovereign control. These efforts aim to create AI systems that can run on cheaper hardware, support public‑sector use cases and reduce dependence on foreign providers. In a broader survey of the region, Rest of World finds that much of Asia remains relatively optimistic about AI, seeing it as a tool for growth and problem‑solving despite global concerns about risk and regulation. From Indonesia to Vietnam, developers are exploring use cases in education, logistics and small‑business support, often under tighter resource constraints than in Silicon Valley.
Reflections
• How might “frugal AI” approaches influence global debates about sustainable and inclusive AI development?
• What can European and US actors learn from Asian experiments with lower‑cost, locally tuned models?

Signal 8.21 – AI as a new tool in cinema
Source
L’IA permet de raconter des histoires qu’on n’osait pas raconter: à Cannes, l’IA comme outil dans le cinéma – Les Échos
Dispatch
Les Échos reports from the Cannes film festival on how directors and producers are experimenting with AI as a creative tool, rather than a replacement for human storytellers. Filmmakers describe using generative tools for concept art, pre‑visualisation, script exploration and subtle visual effects, allowing them to imagine scenes and narratives that would have been too expensive or technically difficult before. Some emphasise that AI can help tell stories that were previously “too risky” or complex to pitch, especially for smaller productions. At the same time, they stress that the core of cinema remains human: actors, writers and directors who decide which stories to tell and how.
Reflections
• How can the film industry use AI to expand creative possibilities without undermining the livelihoods and recognition of human artists?
• What lessons from cinema’s integration of AI can help other creative sectors find a balance between innovation and artistic control?

Signal 8.22 – A president’s warm words for news agencies
Sources
Austrian president trolls Trump, says autocrats attack truth to stay in power – Deccan Herald
Warnung vor betrügerischer Website mit Deepfake des Bundespräsidenten – Austrian Ministry of Finance
Dispatch
In a speach for Minds International, in front of many leaders of the worlds' news agencies, the Austrian president has publicly warned that autocrats attack truth and press freedom to stay in power, stressing that a healthy democracy depends on independent media and verified information. With a remarkable self relativation and sense of humor, he described how he was targeted by a deepfake scam that used his likeness to promote fraudulent crypto investments, prompting official alerts to citizens. Against this backdrop, his warm words about the role of news agencies carry particular weight. He frames agencies as anchors of reliability in an environment flooded with manipulated content, and calls on public authorities to support trustworthy news infrastructures rather than leaving them to fend for themselves against platforms and disinformation.
Reflections
• How can news agencies leverage such high‑level recognition to argue for more sustainable funding and fairer platform relationships?
• How can alliances between news agencies, governments and industries strengthen democratic resilience in an AI‑saturated information space?

Dispatch 8 moves from secret AI talks at the Pentagon and uncontrolled agents in corporate networks to data embassies in the Gulf, gaming studios hiding their use of AI and filmmakers in Cannes exploring new ways of telling stories. Along the way, it shows how existential‑risk narratives can be used as a lobbying tool, how lawmaking is already being reshaped by AI behind the scenes, and how fears of an “AI jobpocalypse” often ignore history and policy. At the same time, it highlights voices that refuse to see journalism as obsolete, that build frugal and sovereign AI stacks in Africa and Asia, and that treat AI as a tool in service of human creativity.
Ending with a handshake in Vienna is a deliberate choice. In a gilded room, the Austrian president speaks about truth, news agencies and the need for verified information in an age of deepfakes and platform power. That scene is a quiet counter‑image to much of the week’s news: against the backdrop of classified deals and global power moves, it reminds us that democratic legitimacy still rests on people who insist that words matter, facts matter and institutions matter.

© APA - Austria Presse Agentur