Of course one should use generative AI, but …

Is it acceptable to use generative AI to write texts? And if so, when and in what way? Are you “authentic” if you let AI generate a text based on your own ideas, and does authenticity even matter? Is it a democratic and inclusive tool that gives everyone a voice, or is it simply “cheating”? Does it impoverish discussion or broaden it? Positions are taken on what is reprehensible and what is permissible. The tone is heated, the moralism evident.

Much less attention is paid to what writing actually is and the function it serves. For the point of writing is not merely to produce text – nor, for some, the exquisite pleasure of finding the right words and arranging them in the right order. The point is also to understand things a little better than one did before. In the process of writing, thoughts are tested, broken down and replaced.

It is through writing that one discovers where one is wrong, where one is unclear and where one lacks sufficient knowledge. It is in writing that opinions change, understanding deepens and new patterns emerge. It is in writing that the obscure becomes clear. As Tegnér might be paraphrased: That which is obscurely written is obscurely thought.

This is why writing is often strenuous and involves resistance. Resistance from the train of thought, resistance from the content, resistance from the words themselves. Filling a page with text is not difficult, but filling a page with the right text may take as long as it takes. The resistance is not inefficiency, however much it may feel like it, but the very mechanism that develops understanding in the writer. This makes resistance essential: writing takes time because it must take time.

We often speak of independent thinking as a right, something contained within the formal recognition of freedom of thought and opinion. In practice, however, independent thinking is more accurately a discipline. It requires repeated effort, resistance and a certain willingness to stand apart. It cannot be handed out as a gift. It must be won, step by step, by each individual.

Freedom of thought and opinion may sound grand, but without independent thinking they amount to very little. In an age where technology offers shortcuts to ready-made answers, independent thinking therefore becomes both more difficult and more decisive.

None of this means that generative AI lacks value. On the contrary. As a tool, it can in many ways improve productivity, structure and clarity. The problem arises when its use shifts from support to substitution.

Cognitive discipline in complex systems

When one hands over all or large parts of writing to generative AI, one does not merely outsource production. One simultaneously outsources parts of one’s own thinking. One gives up practice. One reduces the cognitive effort required to develop analysis and judgement. What one gains in return is, at best, a more polished text and a higher rate of output. What one loses is harder to measure, but undoubtedly more important.

At its core, the mechanism is simple: those who abstain from cognitive effort become worse at it. This applies to understanding. It applies to analysis. It applies to judgement.

Historically, education, research and qualified professional work have been built on the assumption that individuals must develop this discipline themselves. Slowly, and often with frustration. The shortcut that generative AI now offers is naturally attractive, but it creates a structural dilemma: short-term efficiency stands in tension with long-term individual – and societal – capability.

In knowledge-intensive societies, the collective ability to reason, weigh alternatives and make decisions under uncertainty is a critical resource. If this ability gradually weakens, institutions and organisations are affected, and ultimately every political decision. A society in which people cannot formulate their own thoughts and make reasonably independent judgements is a society in which freedom and democracy have been reduced to empty words.

“The most courageous act is still to think for yourself.” – Coco Chanel

The paradox of efficiency and the logic of dependency

It is not difficult to see why generative AI is so appealing. Organisations typically reward rapid output and visible results. Producing more text, more analyses and more supporting material appears rational. Generative AI fits this logic extremely well.

In English, a distinction is drawn between efficiency and effectiveness. The former concerns doing things right – minimising waste of time, resources and effort. The latter concerns doing the right things – actually achieving the goals and outcomes one claims to pursue. Ideally, both are combined. In Swedish, this distinction is lacking, and we often settle for optimising the process and calling it success.

Yet one can be highly efficient in the sense of being fast and productive without coming any closer to what one truly seeks to achieve. And one can reach certain short-term goals while undermining long-term capability. Producing more texts more quickly may appear efficient, but when increased output comes at the cost of reduced understanding, the system becomes more efficient on the surface while its underlying competence – the ability to think independently – is gradually eroded.

Such a development can scarcely be called “success” in any serious sense of the word, regardless of what people choose to label it.

This is not a new phenomenon. Similar dynamics have accompanied other technological shifts, where automation has reduced the need for certain skills while increasing dependence on systems. The difference now is that this concerns not merely manual work, but cognitive processes themselves.

If one writes utilitarian texts – advertising, copy or communications – the consequences may be more limited, as these activities are more concerned with production than with understanding. There is, of course, a risk that writing competence will erode over time, but whether that is a problem is largely a matter of preference. Not everyone wants or needs to be skilled at producing text independently. Thinking, however, I would argue, matters to everyone.

By outsourcing analyses, positions and the development of ideas to generative AI, we outsource our own understanding and judgement. These are extremely difficult to rebuild once weakened. They require experience, effort and exposure to uncertainty. They require mistakes, and the correction of those mistakes. They require time.

Autonomy and intellectual freedom

The ancient Stoics were already clear: no one is free who does not master themselves. In a modern knowledge society, this translates into the ability to think independently. Those who lack this ability are, in practice, never free, regardless of how free the surrounding society may be. Formal freedom of expression, access to information and democratic institutions mean very little if individuals lack the capacity to evaluate, weigh and draw their own conclusions.

Dependence on the analyses of others entails a subtle but real asymmetry of power. Those who do not understand cannot question. Those who cannot evaluate information are at a disadvantage in relation to those who claim to know. Those who cannot formulate their own arguments lose influence to those who can.

Cognitive outsourcing thus creates a divide between those who develop understanding and those who consume ready-made conclusions. The question of generative AI is therefore far larger than matters of efficiency or the importance (according to some) of authenticity when posting on LinkedIn.

Ultimately, it is a question of individual freedom and autonomy, as well as of the very functioning of democratic society.

A choice with consequences

The central question, then, is not whether generative AI should be used. The question is how. If the technology is used to augment human thinking, it can be a powerful tool. If it is used to replace it, it will gradually weaken the intellectual discipline on which modern societies rest.

The issue has technical and organisational dimensions, but at its core it is individual. Systems change only when people change their behaviour. Those who do not take responsibility for their own understanding, who consistently avoid thinking for themselves, should not be surprised when the capacity eventually disappears. The choice is, in principle, free.

But this freedom is formal, not neutral.

Foucault wrote:

“Knowledge is a spiritual adventure and a transformation of the self. The one who knows differs from the one who does not not simply by knowing certain things, but by no longer being the same person. In other words, knowledge is that which transforms the very subjectivity of the knower.

To know, in this sense, is to change. It is a process that requires time, resistance and uncertainty. If we systematically opt out of this process, we also opt out of a part of the person we might otherwise have become.

The question, therefore, is not only what generative AI does to our productivity. The question is who we become when we grow accustomed to not thinking for ourselves.

The choice is free. But not without consequences.

The post The road to unfreedom is paved with good intentions – and generative AI appeared first on Subversivt.

Yet in an era in which the justice system is simultaneously investing in powerful analytical tools capable of mapping connections across vast datasets, a question of legitimacy arises. How is the citizen to understand and relate to the distinction between serious investigative failure, prioritisation, and negligence when it is never clarified?

“Over the course of a single day, a man repeatedly sent me (grossly) sexual images and video clips, of course without my consent. The material was not sent from an anonymous account and the sender was clear and highly identifiable,” writes the politician Jessica Stegrud on Facebook.

Nevertheless, the reasoning provided by the justice authorities reads: “Continued investigation is not expected to lead to the offence being substantiated. It is not possible to prove who committed the act.”

She concludes rhetorically: “Is that reasonable? Is it a sign of a functioning justice system, in your view?”

No, people decidedly do not think so. Her own comment field, and later many others, fill with angry and outraged individuals. What she describes is interpreted as negligence, disinterest, and general incompetence on the part of the police.

That may of course be the case; I do not know what the police have or have not done (I shall return to that problem), but it is not a foregone conclusion based on the information available to us.

What do we know – and what can be proven?

One must distinguish here between what feels self-evident and what one actually knows, and what can be proven in law. The perpetrator is not known simply because Jessica Stegrud or anyone else believes they know who it is.

It is entirely possible to create social media accounts in another person’s name. An existing account can be used unlawfully. Telephone numbers can be spoofed so that a text message or call appears to originate from a particular person when in fact it does not. Moreover, variants of generative AI and techniques such as deepfakes can produce videos and images that appear authentic but are not.

It is therefore insufficient that it appears a certain person is the sender, or that a specific, identifiable person features in the material, for us to know that this individual is the sender. The justice system does not operate on belief but on evidence; even what seems obvious must be investigated and proven.

Sexual harassment as an offence is normally regarded as sufficiently serious to allow the police to request logs and other technical information. It is therefore entirely possible—perhaps even likely—that they did so.

The fact that the case was discontinued does not automatically mean that the police refrained from gathering information. It may mean that the information obtained was insufficient to identify a specific individual.

What are the police actually doing?

When the police request data in such cases, they receive technical information such as IP addresses, timestamps, account names and details of the type of device used. This rarely equates to an identity.

An IP address may lead to a mobile network or a shared address used simultaneously by many people. An operator may sometimes only be able to indicate which subscription had access to the connection at a particular time. VPN services and other anonymisation solutions can render tracing even more difficult or impossible.

Even if the police trace a subscription, it is not self-evident who actually used the phone or computer at the relevant time. Subscriptions may be registered to someone other than the suspect; multiple individuals may have had access to the device. It is therefore not straightforward to link this to a specific person. Something appearing probable is not the same as being provably true.

If there is no supplementary evidence – such as seizure of a device on which the account is logged in, a confession, or other digital traces pointing in the same direction – the investigation often reaches an impasse. One may then have information, but not evidence in the legal sense.

In such circumstances, it is not remarkable if the police assess that it is not possible to prove who committed the act. And if they cannot do so, they will naturally discontinue the case.

That said, none of this demonstrates what the police did, or reasonably ought to have done, in Stegrud’s specific case. Nor do we know how common it is for such data to be requested, or how often the information obtained is sufficient to proceed. For some reason, this is not something the police choose to communicate. Nor do they communicate that this is roughly how their work proceeds and the difficulties they may face.

One may debate whether that is the correct choice, but it undoubtedly leads many people to form views about the police that are neither well-founded nor flattering.

What about prioritisation?

One of the things people “know” when discussing the matter is that unsolved cases are about prioritisation, with perceptions of what is not prioritised and what is prioritised instead. They are at least half right.

The police and the prosecution authority always have more cases than they have resources to investigate fully. This means that even when an offence is formally serious enough to justify certain measures, they must in practice weigh how far it is reasonable to go in each individual case. How much time should be devoted to technical analysis? How many leads should be pursued? How long should a case remain open before concluding that further work is unlikely to yield results?

Two cases may be legally equivalent yet receive different investigative depth depending on workload, available expertise, competing serious crimes and organisational priorities at that particular time.

“Reasonable measures” is therefore very much a relative concept.

Perception, legitimacy and communication

From a systemic perspective, this is central, as it contributes to the arbitrariness perceived by citizens. For the person affected, it does not appear as a question of prioritisation within a portfolio of cases, but as a valuation of their own offence: “Was it not important enough? Was I not worth more effort?”

Even understanding that prioritisation occurs, or accepting it in principle, does not mean one feels secure that the prioritisation is correct in general, or in one’s own specific case.

It is, however, communicatively delicate. If the police were to state openly that a case was discontinued due to (relative) resource constraints, this would likely be perceived as an admission of systemic failure. If they say nothing at all, it risks being interpreted as negligence or indifference, the disadvantages of which are obvious.

In practice, citizens often receive a standard response that explains very little and makes no mention of prioritisation.

When harsh realities are neither articulated nor explained, people form their own interpretations, often negative ones. And when cases spread – as here on social media – further questions arise: what was prioritised instead, and why? Many believe they know, reinforcing the image of a police force more concerned with drink-driving checks and “prosecuting pensioners for hate speech” than with “real crime”.

A closer review of police work, based on available statistics and data, might well demonstrate areas in which public perception is mistaken. But those engaged in heated discussion do not conduct such reviews. They rely on what they believe they see and draw conclusions accordingly. This is facilitated by the fact that the police have left the field open to speculation.

An information vacuum as information strategy

When a case is discontinued with a very brief standard justification, an information vacuum arises. For an outsider, it is impossible to determine whether the investigation was thorough but fruitless, or superficial and swiftly closed.

The lack of information about where in the process matters stalled – or whether there was a substantive process at all – makes it difficult for the individual to decide how to proceed. Should one contest the decision? Seek a review? Complain to an ombudsman? Approach the media? Or accept it? If the police have done everything reasonably expected, it is better to learn to live with the outcome than to expend time and energy contesting it. But if they have not, failing to fight becomes a betrayal of oneself and of the protection the justice system, under the social contract, promises.

For the victim, this often has tangible emotional consequences. Beyond the original violation comes a secondary burden: the feeling of not being taken seriously, of being left alone with the outcome, of suspecting that the justice system did not even try but simply moved on.

The lack of transparency also shifts the burden onto the individual victim to have the strength to challenge the decision. Those without resources, knowledge or stamina fall away. That is hardly satisfactory for authorities expected, according to every policy document and value statement, to treat citizens equally.

There is also a clear asymmetry of cost. For the justice system, a discontinued case is one decision among many. For the victim, it may be the endpoint of a long and distressing process.

This opacity therefore has consequences – not only for the individual victim but for trust in the justice system as a whole. When people cannot distinguish between a serious investigative failure and mere passivity, they tend to interpret matters in the worst possible light, especially where trust is already low.

Responsibility and transparency

Ironically, it is therefore the police’s own choice not to render their decisions more comprehensible that causes good work to be perceived as bad, and trust to erode as a result. It is the police, not the citizen, who have actively chosen not to be more transparent about their processes and priorities.

Ultimately, however, citizens pay the price. However one turns the matter, both individuals and the collective lose when they lack confidence in the justice system to which they have ceded their right to protect themselves.

All this would be troubling enough in itself. But the picture becomes considerably more problematic when one simultaneously considers the tools the police in fact use.

Great power. Very little insight.

The police are not necessarily as helpless in the face of technical difficulties as my initial description suggested. While discontinued investigations are often justified by stating that it cannot be proven who committed the act, that the evidential chain does not hold, or that the link between technical information and an individual is insufficiently strong, the police use one of the world’s most advanced systems for large-scale data integration, analysis and pattern recognition: a system designed precisely to connect fragmented information into coherent decision-making material, specifically developed for intelligence and surveillance purposes—namely Palantir Gotham.

Palantir Technologies’s Gotham platform enables the linking, visualisation and analysis of vast quantities of data. Concretely, information previously dispersed across different registers, reports and tips can be automatically connected and analysed as a whole. The system can display connections between persons, telephone numbers, vehicles, addresses and events, presenting them as network maps, timelines or geographical overviews in which patterns become visually apparent. It is not merely a matter of storing large volumes of data, but of identifying relationships that might otherwise be difficult or impossible for an individual case officer to detect.

In practice, the system may identify that a person appearing in a traffic stop also has links to an address associated with suspected activity, that the same telephone number recurs in multiple investigations, or that a particular pattern of events repeats in different parts of the country – or, at least in theory, map in detail a person suspected of sexual harassment.

AI and machine learning are central components of Gotham. Through Palantir’s AIP platform, generative and analytical AI are integrated into existing data processes, enabling predictive analysis, pattern identification and automated decision support.

This means the system does not merely show what has already occurred, but can assist in assessing what is likely to occur. Predictive analysis may, for example, identify areas at heightened risk of certain crimes based on historical patterns. Pattern recognition can detect anomalous behaviour – such as unusual transaction flows or communication patterns – warranting closer scrutiny.

Automated decision support implies that the system may suggest priorities or measures – such as discontinuing a case—based on aggregated data, although formal decisions must legally be taken by human officials.

To suggest that such a system could not succeed where more traditional systems fail strains credulity, whatever the police’s practical use of it.

Technical capability and political consequence

When the police employ systems enabling broad analysis of extensive datasets, yet simultaneously inform victims that “it cannot be proven who did what”, a crisis of legitimacy arises.

We face two principal possibilities: either Palantir’s capacity is utilised fully – raising serious questions about proportionality and legality – or it is used selectively and restrictively, prompting the question why such a controversial and powerful tool was procured at significant expense.

Palantir Gotham is about as far from a neutral technological choice as one can come.

Technology on a collision course with the rule of law

Swedish law rests upon clear principles: legality, proportionality, purpose limitation and transparency. Coercive measures and intrusions into personal integrity must be proportionate to the seriousness of the offence and the concrete investigative need. Data collection must be necessary, specific and controllable. These principles are expressed in procedural law, police legislation, public access and secrecy legislation, the GDPR and the EU AI Act.

Palantir’s systems, by contrast, are designed for a different paradigm: maximal data integration, broad cross-access to linked information and analysis at population level. Criticism of Palantir in other European countries has therefore centred precisely on legal and democratic oversight.

Palantir is a US company subject to the Cloud Act, which in certain circumstances may compel disclosure of data to US authorities even where the data are stored in Europe. This stands in direct tension with Swedish and European law and cannot be resolved merely by contractual arrangements.

There is also the issue of opacity. The source code is not open. The algorithms function as a black box. There is no external scrutiny of how data are weighted, how patterns are identified or how risk assessments are conducted. For Swedish authorities, this poses real challenges in relation to the GDPR’s transparency requirements and the AI Act’s demands for documentation, traceability and verifiable reliability in high-risk AI systems.

Constitutional conflicts in practice

In Germany, this has led to open conflict. In 2023, the Federal Constitutional Court held the use of Palantir Gotham unconstitutional, particularly criticising uncontrolled data integration from multiple sources, insufficient transparency regarding whose data were analysed, and inadequate safeguards against profiling individuals not suspected of crime.

What was deemed unlawful was, concretely, that individuals not suspected of any offence could have their data compiled, analysed and assessed by police without their knowledge, without the opportunity to object, and without clear legal limits. The state acquires a tool that sees connections everywhere, while the individual has no insight into when, how or why they have entered the system.

How is an authority unable fully to explain how a system functions to demonstrate that data processing has been necessary, proportionate and correct? And if the system is used in decisions to discontinue cases, how can the reasoning be explained to citizens?

Cost, dependency and digital sovereignty

Palantir licences are also extremely expensive. Adapting the system to European and national law often requires key functions to be disabled or restricted. German experience indicates that substantial functionality remains unused while IT departments are heavily burdened. Exit strategies are often lacking, creating vendor dependency difficult to reconcile with long-term public administration – an issue sharpened by increasing awareness of digital sovereignty. After five years of use, could the police change system even if they wished?

The political dimension

Another problematic aspect concerns Palantir’s political and ideological context. The company’s co-founder and principal owner is Peter Thiel. He has not only expressed far-reaching criticism of liberal democracy, but has also argued in favour of elite-driven societal models in which popular influence plays a subordinate role – a view of society in which political power ought to be concentrated in a technological and economic elite rather than legitimised through broad democratic participation.

It is therefore hardly surprising that he is often described as an almost archetypal “Bond villain”: an extraordinarily wealthy technology entrepreneur with openly system-critical ambitions and direct influence over political power. Thiel is a key financier behind Donald Trump and JD Vance, and has also funded candidates and networks linked to national-conservative movements in the United States.

This affects public acceptance – and thereby the legitimacy – of its use within a Swedish rule-of-law state. In the current geopolitical climate, the matter is far from straightforward. Given the prevailing situation, do we wish to allocate more of our taxpayers’ money to someone who supports Trump with substantial financial contributions?

It is, of course, possible to debate how such questions ought to be approached – both in general and in this particular instance – but at some point the public sector should ask itself which organisations it is supporting, directly and indirectly, and what the long-term consequences may be. Sweden does not exist in a vacuum; recent statements and developments in the United States have undeniably affected us, for example.

There are therefore many question marks surrounding the use of Palantir.

For citizens, the cumulative result is ambiguity and contradiction, which in the long run erodes their trust. If the state collects, analyses and cross-references data on a large scale, yet still cannot – or will not – explain how it operates and why it fails, trust is undermined. And without trust, neither legal certainty nor the rule of law functions effectively in practice.

The importance of transparency

The justice system is rightly bound by high standards of proof; it must not operate on suspicion, intuition or social consensus. It is also bound by prioritisation and resource constraints, which are practically unavoidable. It is increasingly dependent on complex, opaque technical systems that are difficult to explain – even to their users.

None of this need necessarily be problematic in itself. But when these factors combine with an almost total absence of intelligible feedback to the individual citizen, a serious problem arises.

For the victim, the entire process is reduced to a brief statement: it cannot be proven. What was tested, what was weighed, what was possible but deprioritised – all of this disappears. What remains is the sense that the state either did not care or could not cope. In practice, it scarcely matters which is true, as the experience is the same.

In Stegrud’s case, it is entirely possible that the police did everything reasonably possible. It is also entirely possible that they did not. The point is that we do not know. And that uncertainty is not an unfortunate by-product – it is a consequence of how the system has chosen to organise its work and its communication.

This is the core of the problem: not that the justice system sometimes fails, but that it does not explain its failures. Not that all offences are not solved, but that it is impossible to determine whether they were even examined in a manner proportionate to their seriousness.

Trust is not built on the state always succeeding, but on its being comprehensible, reviewable and, to some extent, accountable. When that possibility disappears, trust is replaced by speculation, and legal certainty by suspicion.

In that sense, increased transparency is not a benevolent concession by the police but a systemic necessity. Only when citizens can distinguish one thing from another can they accept that the state sometimes falls short. And that, in turn, is a precondition for maintaining the trust upon which legitimacy rests.

Jessica Stegrud’s question – Is it reasonable? Is it a sign of a functioning justice system? – cannot be answered with a simple yes or no. But so long as the system makes it impossible to understand why a case is discontinued, the answer for more and more people will be no. And that is a far greater problem than any single discontinued case.

The post “It is not possible to prove who committed the act” – on standards of proof, Palantir and the silence of the state appeared first on Subversivt.

“Do you want to make a difference to society and contribute to Sweden’s independence in the world? We are now looking for an Azure technician with a focus on the development, operation and management of FRA’s cloud solutions. Does this sound like something for you? Apply now!”

This is the wording of an advertisement issued by the National Defence Radio Establishment (FRA). The FRA is a Swedish intelligence agency under the Ministry of Defence, tasked with collecting and analysing signals in electronic communications.

Organisational infrastructure is rarely as well protected as one might imagine. Through open sources – such as job advertisements, technical requirement profiles and procurement documents – it is entirely possible to gradually construct a clear picture. Each individual detail may be harmless in isolation, but their combined effect is far more revealing.

When a public authority requests expertise in specific technological ecosystems, such as cloud platforms or vendor-specific identity solutions, it becomes possible to make reasonable assumptions about dependencies and architecture, for example. This may not reveal everything in exact terms, but one does not need complete information in order to conduct strategic analysis.

Thus, organisations’ own requirements for transparency – or simple carelessness, for that matter – create opportunities for infrastructural mapping. This does not, of course, mean that classified information has been, or will be, exposed. But it does make matters considerably easier for anyone with such intentions, all else being equal.

Microsoft is subject to US jurisdiction

If we assume that Azure is being used by a security-conscious authority, we must also assume that such use is restricted. That it concerns support functions, development environments or application hosting without any interaction with classified information, rather than the cloud storage of such data.

This does not, however, change the fact that Microsoft (which provides Azure) is subject to US jurisdiction. This means that political decisions in the United States can affect the room for manoeuvre of a Swedish authority, regardless of where data is physically stored or how well it is encrypted.

Consequences of non-compliance

In the event of a political conflict between the United States and Sweden – for example, if one party seeks control over Greenland – consequences may arise both with and without active coercive measures. Uncertainty regarding future access, changes in contractual terms or limited support all create a structural vulnerability.

If we further assume that the authority has a technical dependency on Azure (or other Microsoft services), this can, of course, be used as a means of political pressure. The ability to rapidly replace or dismantle such dependencies is limited, particularly in terms of time.

I do not wish to believe that the FRA stores classified data via Azure. Nevertheless, it should be noted that the fact that Azure has data centres for cloud storage in Sweden does not alter the ability of the US authorities to gain access to that data under the Cloud Act.

Without digital sovereignty, there is no sovereignty

Open sources thus enable reasonable assumptions about infrastructure and vendor lock-in, while legal and political control over those vendors creates risks that simply cannot be mitigated through technical safeguards alone.

It is difficult not to note the irony that this advertisement appears on the very same day that Trump openly expresses his dissatisfaction with Sweden and threatens punitive tariffs in order to force compliance in the Greenland issue.

What will be the next step if we do not comply?

The post Sovereignty as a Service – on the Security Policy Implications of Frictionless Technology Choices appeared first on Subversivt.

In May 2025, Stockholm City Council rejected a request from the US Embassy that the city dismantle its diversity, equity and inclusion (DEI) programmes.

I do not know precisely which applications Stockholm uses or where its data is stored. However, it is far from unlikely that, like many public-sector organisations, it has become dependent on American companies and services such as Microsoft, AWS and Google.

Stockholm’s refusal to comply with the embassy’s request could therefore, in theory, have led to those companies receiving instructions from US authorities to restrict or disable certain services on national security grounds. Such measures are entirely possible under US law.

That did not happen. But the scenario I am describing is neither implausible nor far-fetched.

Because it has already happened.

The most recent example is Microsoft’s decision to block the email account of the Chief Prosecutor of the International Criminal Court in The Hague following sanctions imposed by the Trump administration. His bank account in the United Kingdom was also frozen.

For Sweden, this vulnerability extends to large parts of the country’s digital infrastructure: the systems that enable public institutions to function. It includes school platforms, healthcare records systems, financial administration, social services data and municipal budgeting. It also encompasses everyday workplace applications such as Microsoft Outlook and Teams.

In short, the current US administration possesses the practical ability to disrupt large parts of Swedish public administration if it chooses to do so. Such an action would make the 2024 ransomware attack on Tietoevry seem relatively minor by comparison.

This represents a far greater threat to the functioning of society—and, I would argue, a far more plausible one—than the prospect of Trump ordering troops onto Swedish territory.

(Although that judgement admittedly rests on the assumption that his actions remain rational and reasonably predictable, an assumption that may itself be questionable.)

This is therefore not primarily a technological issue. It is a matter of national security.

Sweden finds itself in this position because it has prioritised operational convenience over secure and independent operation. That approach will work perfectly—right up to the point at which it no longer does.

And that point could arrive at any moment, particularly given the volatility of the current US administration.

It is remarkable, though perhaps not surprising, that the Swedish government, along with much of the public sector, continues to bury its head in the sand.

Further Reading

Opinion: Europe Needs Digital Sovereignty

“The United States can instruct its technology giants to withdraw from Denmark unless Greenland is handed over, effectively paralysing Danish society. Today, there is no meaningful distinction between digital sovereignty and sovereignty itself.”

Read the full article here: Europe Needs Digital Sovereignty

The post Swedish Public Administration in American Hands appeared first on Subversivt.