🧠Editorial:

Sometimes you only realize how dependent you are when something fails. Or is sold. Or suddenly someone else wants access to your data.

That’s exactly what this issue of my Weekly Backlog is about.

The USA wants access to European police data. Billions of images from Pokémon Go become training material for AI years later. An Exchange outage reminds companies how much of their communication relies on a single provider. And AWS acquires the company behind DuckDB – while the code remains Open Source.

What particularly stuck with me this week is the question of how decisions made today sometimes reveal their true significance only years later.

Those who walked through the city with Pokémon Go in 2016 probably didn’t think: I’m collecting training data for an AI that will eventually help robots navigate.

Those who move all their communication to Microsoft probably don’t think about what happens if central authentication fails with every email.

And while an Open Source license answers the question of who owns the code, it doesn’t determine who shapes the future of a project.

This week also brings a rather wild Linux approach without a classic hypervisor, Kubernetes 1.37, and some interesting stories about openDesk, Microsoft dependencies, and cybersecurity.

My backlog was definitely much longer than my available time this week. Enjoy reading. ☕️

📰 Tech-News:

European Police Data for the USA

The USA wants extensive access to European police databases. Those who don’t comply risk losing visa-free travel for their citizens. The EU Commission is negotiating a corresponding framework agreement.

I already find this highly problematic. It becomes even more problematic with a view to Palantir.

The current draft of the “Enhanced Border Security Partnership” allows automated queries of European police databases. The data may also be used for “screening” individuals. Under certain conditions, information about political opinions, religion, ethnic origin, health, or sexual life can be transmitted.

And these data meet a security infrastructure in the USA where Palantir already plays a significant technological role.

Palantir is not just any software provider for me. Co-founder Peter Thiel wrote back in 2009 that he no longer believes that freedom and democracy are compatible. CEO Alex Karp publicly advocates a political agenda in which the technological and military power of the USA plays a central role.

The agreement does not mean that Palantir automatically gains access to European police data. But Europe would significantly expand data transfer into a US security ecosystem where Palantir is firmly anchored. At the same time, the USA lacks the independent data protection oversight that European fundamental rights demand. Max Schrems considers the planned model illegal.

And all this so that Europeans can continue to travel to the USA without a visa.

I believe: European fundamental rights should not be bargaining chips for American entry conditions. And European police data do not belong in a “security architecture” of Palantir.

🔗 https://netzpolitik.org/2026/erzwungene-grenzpartnerschaft-abkommen-zum-us-datenzugriff-auf-der-zielgeraden/

Your Data from Yesterday Trains Tomorrow’s AI

Millions of people walked through cities with Pokémon Go, pointed their smartphones at buildings and landmarks, and searched for virtual figures. What was incidentally created now takes on a completely new meaning.

Niantic Spatial has trained its model with 30 billion images from urban environments. The recordings come from players of Pokémon Go and Ingress, among others. They contain detailed metadata: location, direction, orientation of the smartphone, movement, speed, and other information.

With this, a model today can determine its location based on a few images – according to Niantic Spatial, accurate to within a few centimeters. Delivery robots are expected to navigate where GPS reaches its limits.

Technologically, this is impressive.

But that’s exactly why we should talk about the other side.

Pokémon Go was perceived as a game. It was played by many young people. At the same time, its use created a dataset of enormous economic and technological value.

This again highlights a fundamental rule of the digital economy:

Nothing is free.

If we don’t pay money for a digital service, it doesn’t mean there’s no exchange. We may be paying with location data, movement profiles, images, usage behavior, or information about our surroundings.

And the crucial point is: At the time of data collection, we often can’t estimate the value these data will have years later and what they can be used for technically.

Data that originated in the context of a game becomes training material for AI. Smartphone recordings become a highly precise digital representation of our physical world. And a leisure product becomes a building block of commercial infrastructure.

This is not automatically illegitimate. But it shows why we need to talk much more consistently in Europe about data sovereignty, purpose limitation, and digital sovereignty.

Data are not a worthless byproduct of our digital activities.

They are an asset. And too often we only recognize their value after we’ve given it away.

🔗 https://t3n.de/news/pokemon-go-ki-lieferroboter-navigation-1733967/

Microsoft Exchange Fails. And Suddenly Digital Dependency is Back on the Table.

On Monday, there were significant issues with Microsoft Exchange Online. Users were partially unable to send or receive emails, had problems logging in and accessing their mailboxes. Search functions and administrative functions were also affected.

Microsoft confirmed the incident and attributed it to a problem with a central authentication configuration used by several Microsoft 365 services. Exchange Online was particularly affected.

According to Microsoft, the incident began on August 31 at 15:08 UTC. As it progressed, the company reported increasing restoration of mail connectivity. However, queued messages could still cause delays for individual organizations.

Exchange Online is a central communication infrastructure for numerous companies and organizations. When a service of this magnitude has problems, it doesn’t just affect some software function. It affects an essential part of the workday.

This is precisely where the structural problem of increasing cloud centralization becomes apparent.

The more email, collaboration, identity management, office applications, and other business-critical processes converge at a single provider, the greater its significance for one’s ability to act.

Cloud is not automatically resilience. And a large provider does not eliminate dependencies. It can even concentrate them.

Digital sovereignty therefore does not mean rejecting every cloud service. It means knowing dependencies, having alternatives, and designing infrastructure so that the failure of a single provider does not lead to a loss of control.

The Exchange outage will be resolved - the structural dependency remains.

🔗https://www.heise.de/news/Ausfall-von-Exchange-Online-Microsoft-untersucht-den-Vorfall-11435865.html?utm_term=Autofeed&utm_medium=Social&utm_source=LinkedIn#Echobox=1788200605

Linux Without Hypervisor: Multiple Kernels to Run Parallel on a Single Machine

Multiple Linux kernels running parallel on the same hardware – without classic virtualization through a hypervisor. That’s exactly what long-time kernel developer Cong Wang is working on with MKLinux.

The idea of an operating system kernel per processor core is not new. Researchers from Microsoft and ETH Zurich were already exploring this approach 17 years ago. In practice, however, hardware virtualization using a hypervisor prevailed.

MKLinux takes a different path.

First, a kernel starts that manages all the hardware. It can then release parts of the hardware and assign them to additional kernels started via kexec. These receive direct access to the hardware allocated to them – even without Single Root IO Virtualisation (SR-IOV).

Additional hardware can be assigned later via hot plugging. For the initial allocation, MKLinux uses the Device Tree. The concept thus builds on existing technologies.

Wang sees advantages over existing isolation mechanisms. Like containers, software runs without the abstraction layer of a hypervisor. Hypervisor calls and additional page tables are eliminated. Benchmarks published by Wang show significantly lower latencies in some cases for system calls and context switches.

At the same time, the individual kernels are isolated from each other. A problematic kernel should not be able to bring down the entire system. The separation can also limit the attack surface.

Wang cites hosting different server components on a single machine as a possible use case. The concept can also be extended to individual applications: A device kernel manages the hardware and starts additional app kernels. Communication takes place via shared memory. For this approach, Wang developed a dedicated file system with Daxfs. He sees app kernels as a potential solution for AI agents.

An initial version of MKLinux has already been released. The modified Linux kernel 7.0 is available on GitHub.

Currently, MKLinux only supports 64-bit x86 systems. Whether the concept will be included in the mainline Linux kernel in the future is open. A short-term integration is not expected.

🔗https://www.golem.de/news/ohne-hypervisor-linux-fork-startet-mehrere-kernels-auf-einem-computer-2608-212438.html

💬 LinkedIn Post of the Week:

AWS Buys DuckLabs

DuckDB remains Open Source, MIT-licensed, and with the non-profit DuckDB Foundation. However, the company behind the project and the people who have significantly advanced its development are moving to AWS.

Dominik Rapacki uses the acquisition as an opportunity to highlight an often overlooked aspect of Open Source: A free license regulates what can be done with the code. But it does not decide who will continue to develop it, set priorities, and thus shape the direction of a project.

Particularly interesting is the European story behind DuckDB. The project originated from publicly funded research at CWI in Amsterdam, was released as Open Source, and later spun off into DuckLabs. Now AWS takes over the company.

Dominik Rapacki’s post contextualizes this path and shows why in Open Source, the focus should not only be on licenses but also on the people and structures behind a project.

🔗 https://lnkd.in/p/ecTVZkJb

☸️Kubernetes-News:

Kubernetes 1.37: Metrics API Reaches Stable Status

With Kubernetes 1.37, the Resource Metrics API metrics.k8s.io is released as v1 after almost nine years in beta. Used by kubectl top and the HorizontalPodAutoscaler, among others.