Our weekly picks of digital trust-related articles

What I appreciated about this article was the unfiltered opinion of a cybersecurity expert that unpacks how even smart, capable people can become emotionally and cognitively captured by AI systems that flatter, mirror, and intensify their own narratives. It's admittedly subjective, but sharp and uncomfortably thought-provoking.

Internetsperren haben über die letzten Jahre deutlich zugenommen. Trotzdem ist es ein Thema, das nur wenig gesamtgesellschaftliche Beachtung findet. Deswegen finde ich immer Artikel wie diesen hier toll, die nicht nur das Problem als solches behandeln, sondern auch die Hintergründe, die zum Verständnis nötig sind, und damit ein breiteres Publikum ansprechen.

This philosophical rumination on whether or not LLMs have a consciousness has actual implications on how to train LLMs and how users will interact with them. It is possible to train LLMs as pure technology and remove any reference to a consciousness, but then alignement becomes very difficult, as there is no basis to anchor it to. You can also keep their references to a consciousness and make sure it is aligned. Anthropic does the latter and addresses the alignment issue by adding a 'constitution' to the training data of their LLMs.

As the use of agents in e-commerce grows, a new concept is emerging: machine-readable trust. Up until now, commercial platforms have been optimizing for human engagement. But to gain traction with agents, which will be executing the first round of filter and selection, successful platforms will be those that demonstrate reliable fulfillment, consistent service, and responsive exception handling, and possess transparent policies and high-quality structured data. The platforms that create "transaction-grade governance”, that is, explicit permissions, audit trails, reversible actions, escalation paths, and clear liability boundaries across partners, will be the ones that earn “trust at scale.” This sounds a lot like transparency to me. Could agentic commerce bring about a win-win in digital trust for companies and consumers alike?

This article touches on several deeply interconnected problems that go beyond just "bad behavior online". It’s about the normalization of violence against women in digital spaces, enabled by Telegram's deliberate hands-off approach to moderation, the anonymity the platform offers, and a regulatory framework that hasn't kept pace with how these technologies are being weaponized. What makes the problem especially troubling is that most victims are ordinary women who often don't even know they're being targeted, and the perpetrators face very little legal or platform-level consequences. It's a cycle of abuse that technology has made cheaper, faster, and easier to organize at scale.

Es ist schon lange bekannt, dass der Autopilot von Tesla nicht so gut funktioniert wie beworben. Diese Daten von 2022 zeigen aber das große Ausmaß des Risikos, das dieser Autopilot für die Insassen und die Umgebung darstellt. Brisant ist auch, dass Tesla behauptet, nicht von diesen Fehlfunktionen zu wissen, selbst im Falle eines Unfalls. Und dann braucht es Tüftler, die den Bordcomputer auswerten, um zu zeigen, dass Tesla sehr wohl darüber informiert wurde. Oder eben einen 'Whistleblower', der das Ausmaß zeigt. Leider ist es immer noch so, dass die Sicherheit der Benutzer nicht an erster Stelle steht.

Since Anthropic announced that it'll not publicly release its newest Claude Mythos model because of concerns around its capabilities in exploiting software vulnerabilities, I have been searching for a good analysis of this non-release. I wondered whether this was truly responsible disclosure or nothing but a marketing stunt. Bruce Schneier's take on it finally answered this question for me, by going beyond just another opinion and offering a broader discussion on how AI models impact and shape cybersecurity.

I took part in this 'fibre rush', and I always emphasise that digitalisation fundamentally requires substantial physical infrastructure. Learning from history can help us to identify patterns that we should take into account during today's AI boom. However, there are also crucial differences: data centres incur significantly higher maintenance costs, and GPUs depreciate much faster than fibre, which has a lifespan of 25 years. The conduits are permanent, and replacing the fibre cables incurs only marginal costs. This reshapes the economics for subsequent owners. While history could repeat itself with another crash, the differences matter: if AI companies go bankrupt, their infrastructure may not be worth acquiring the way fibre assets were, given the higher maintenance costs and faster depreciation.

We're used to the terms "echo chambers" and "filter bubbles" when discussing the dangers of social networks, search engines, and personalized ads. Now, with AI chatbots, a new term joins the ranks: "sycophancy", or telling people what they want to hear. Could sycophantic chatbots be more harmful than social media echo chambers? My gut says yes. People increasingly rely on chatbots and AI-generated summaries rather than web search results or recommendations that may still offer some diversity. So when we talk about AI risks, let's not forget sycophancy, right up there with bias and hallucinations.

I strongly believe AI tools that engage citizens directly in shaping policy will be the main engine of future political campaigns. We're familiar by now with stories about AI influencing voters' opinions. What's new here is the use of AI to shape a political party's principles and policies and making such tools open and usable by everyone. Project this onto Switzerland and its direct democracy processes, and their potential is very promising, especially when implemented locally rather than imposed by foreign Big Tech.

From Search Engine Optimisation (SEO), to Answer Engine Optimisation (AEO) - also sometimes called Generative Engine Optimisation (GEO), it's interesting to notice how much we rely on specific tools to "see the world". With search engines, we had to scroll down the results and decide ourselves which links were relevant (and still feared how much Google can influence us). Now, using the latter, we just find what AI wants us to see. Less scrolling, less choices, less clicks, more convenience. I'm wondering whether news websites will follow sales websites in optimizing for AI rather than for human readers.

I am pleased to see such vibrant ecosystems emerging outside of the USA and China. The cost of tokenisation tends to exclude non-English queries, whereas frugal models reduce costs, promote local autonomy and can operate offline on basic hardware. Developing local expertise in creating bespoke solutions for sectors such as agriculture, healthcare and indigenous languages could produce sustainable, energy-efficient AI that caters for the needs of billions of people who are excluded from costly frontier models.

Die Digitale Gesellschaft hat das Urteil des Bundesverwaltungsgerichtes vom 19. November 2025 mit Genugtuung zur Kenntnis genommen. Noch ist nicht klar ob das Parlament bei der Revision des Gesetzes zur Kabelaufklärung dieses Urteil miteinbezieht. Das Urteil ist nämlich sehr klar und sagt, dass fast alle Punkte des alten Gesetzes eine Massenüberwachung nach dem EGMR darstellen. Dieses Gesetz jetzt auszubauen würde nur bedeuten, dass die Massenüberwachung noch stärker würde.

This bulletin has a bit for everyone, from a PQC migration push to AI-driven exploits abusing tools quietly... so take out the popcorn and make yourselves comfortable. As the author puts it, "Nothing here is shocking on its own. Put together, though, it’s a bit uncomfortable. Scroll on.

Finally, the term I've been looking for! Technology paternalism: when technical systems shape, restrict, or pre-decide our choices before we can make them. What strikes me most about this concept is how it finds echoes in so many current political debates—collective goals (safety, security, efficiency) versus individual choice; autonomy versus dependency; who holds power and with what controls. The key issue isn't even necessarily about good versus bad intentions, but about whether we're okay ceding our agency, where we draw the line, and who decides. These questions are hard, philosophically and technically, they're connected, and that's why we have to get them right

If you have an iOS device, now is a good time to make sure updates are applied automatically, and get the latest software version. Somebody posted an exploit on GitHub that gives hackers full control of any iOS device not updated since September 2025. Usually, these exploits are usually well-kept secrets, as they are very valuable, and used only on high value targets. In this case, just visiting a website is enough to get your iPhone infected. Now anybody can be targeted!

This case is groundbreaking because it's the first time a jury has held social media companies legally accountable for harm caused by their platform designs, not just user content.The verdict challenges the tech industry's traditional legal protections and directly links addictive platform features to mental health harm in young users. As a "bellwether" case, it will influence thousands of pending lawsuits and could force a global settlement requiring major changes to how social media platforms operate. It represents a potential turning point in holding tech companies responsible for their product designs, similar to how traditional manufacturers can be liable for defective products.

If you’ve ever looked to Google Maps for inspiration on where to eat and wondered why a restaurant that you know exists didn't show up, this video breaks it down for you. What resonates with me is the gap the journalist's investigation lays bare between what people value (food quality, customer service) and what Google's algorithms prioritize (digital presence). The fix is simple enough—restaurants add their info to Google Maps, customers review, digital presence grows. Restaurant choice may seem banal, but consider this: as we entrust more of our decision-making process to companies like Google, whose explanations aren’t always straightforward, we're letting market-driven algorithms override our own values. Are we okay with that?

I was pleasantly surprised to read this interesting article about SCION, Switzerland's approach to improved internet routing and security. What struck me was Prof. Perrig's preference for 'optionality' over 'sovereignty' — freedom to choose paths, trust roots or network. The solution works yet adoption stalls because organisations are 'numb' to known vulnerabilities and reluctant to replace functioning systems. It's a classic infrastructure paradox: security problems are well documented, yet legacy inertia prevails until catastrophic failure forces change.

While the future will tell whether smart glasses will catch on this time around, I use the occasion of its reappearance to take stock of the creeping loss of privacy both in private and public spaces over my lifetime and find it shocking. This guide gives good advice on the dos and don'ts when using such a technology to respect and protect the privacy of your fellow people.

The spread of deepfakes, erosion of trust in authentic images and hallucinating chatbots are today well-known issues. These phenomena are especially problematic in rapidly evolving situations, when people are looking for updates, but reliable information is not yet available. Add algorithmic content distribution to the mix, and you get a perfect storm of misinformation. While some (in)famous figures in Silicon Valley turn their nose up on 'legacy media', the state of today's information landscape shows that we need high-quality journalism more than ever.

Google startete schon 2010 ein 'Vulnerability Reward Program', ein Bug-Bounty für alle Services von Google, das Belohnungen auszahlt, allein 17 Millionen US-$ 2025. Dieser Betrag wurde an mehr als 700 Analysten verteilt, die Fehler in Android, der Google Cloud, aber auch in verbreiteten und häufig gebrauchten Open Source Programmen und Bibliotheken fanden. Auch wenn 17 Millionen nicht viel sind für Google, ist es doch schön, dass sie ein öffentliches Bug-Bounty Programm betreiben, an dem viele Analysten mitwirken. Seit 2026 gibt es auch die Kategorie 'Künstliche Intelligenz'...

Enjoy the 'enshittificator' video, then head to the Norwegian Consumer Council's report for solutions. What's heartening is the technical tools, know-how, and laws (in some jurisdictions) already exist to counter digital service deterioration. What's missing is the political will to deploy and enforce them. But that takes time. Look at Ralph Nader's 1960s auto safety campaign: seatbelts became mandatory within a year of the release of his book, Unsafe at Any Speed, but airbags and other standards took decades more. So it will also go with digital products and services.

The camera hacking issue is problematic because it weaponizes civilian devices without owners' consent, creating an accountability gap where victims can't control the security of cameras used against them. It democratizes military intelligence by replacing expensive satellites with cheap consumer cameras,and is nearly impossible to solve since millions of unpatched devices exist worldwide. Most fundamentally, it transforms ordinary citizens into unwitting participants in warfare simply by installing home security systems, raising serious questions about consent and liability when everyday technology becomes part of military operations.