Our weekly picks of digital trust-related articles

This makes a refreshing change from the usual LLM apocalypse: your data is still valuable, especially when there is so much of it! Once again it shows that collecting and storing data long-term is a bad idea. There needs to be strict control on how long personally identifiable information (PII) can be retained, after which it must be deleted. Compared to the publicly available data on LinkedIn, this data source is of much higher value: it is verified, includes complete addresses, and captures people who have never been online at all.

Yet another push to ban social media for minors under 15, this time EU-wide, via the forthcoming EU Kids Act. What is striking is how little the announcement engages with the framework that already exists. The EU GDPR already requires parental consent before minors' data can be processed, with the age threshold set between 13 and 16 depending on the member state. And the EU DSA already obliges platforms to put child-protection measures in place. In other words, the legal scaffolding for protecting minors online is already here. The gap is enforcement, not legislation.

Barely registering above the din of AI doomsayers are the voices reminding the public that legal and regulatory mechanisms, like consumer protection laws and anti-competition regimes, already exist to address many of the damages and harms caused by AI. The article offers hypotheses for why this line of conversation isn’t more front and center. But the more important question is how do we create the political will in citizens and their representatives to come together and use these legal instruments to hold frontier labs accountable.

If you think cases like these are regrettable exceptions, think again. With technology permeating more and more of our daily lives, it has become a go-to tool for domestic abusers. We often talk about the security and trustworthiness of these systems in very sophisticated scenarios, yet forget that a less tech-savvy person relinquishing control over them to their partner can lead to a very real danger for many as well. We urgently need to include these considerations into our evaluations of the societal impacts of new technologies.

This is a lighthearted article about formal verification in software and mathematics. Formal verification takes a mathematical statement or a software program as input. In this article, that statement is Fermat's last theorem: a^n + b^n != c^n. You then add the constraints you want verified, in this case that all numbers must be natural numbers greater than 0, and that n must be greater than 2. Lean, the formal verification program, then tries to construct a proof of why this is true. For Fermat's last theorem, the proof is usually very long, but due to a bug in Lean, you can produce a surprisingly short one. This shows that trusting provers is not always straightforward, and that the provers themselves need to be verified!

I'm following the discussion on whom to blame for LLM attacks on our internet infrastructure. I found the following article which summarises the debate nicely, even though I don't completely agree on who should be doing the work. The article compares three approaches to responsibility: state regulation, product liability, and electronic personhood. But all fail to correctly grasp the new challenge these attacks present. My personal take is that the government should make it clear that the companies creating these tools are responsible for the resulting hacks and must be held accountable.

Plus d'une vingtaine de signalements reçus en quelques mois par l'association française e-Enfance concernant des femmes filmées à leur insu par les lunettes connectées Ray-Ban Meta. Ce chiffre, assez bas en apparence, doit être remis dans son contexte : la plupart des victimes ne se rendent pas compte qu'elles ont été enregistrées et une autre majorité d'entre elles se sentent démunies et ne savent pas forcément comment agir ! Ce chiffre doit donc nous inquiéter et nous faire réfléchir sur la façon dont nous souhaitons aborder ces technologies intrusives. Pour aller plus loin, vous pouvez écouter le podcast Le Point J dans lequel j'ai répondu aux questions de Jessica Vial.

The genie metaphor is compelling. However, a folklore genie is mischievous: it deliberately exploits the gap between what you said and what you meant. AI agents aren't gaming your instructions; they're simply executing an interpretation without knowing it might be wrong. The problem isn't misunderstanding, it's misinterpreting your intentions. A human flags that uncertainty. AI agents don't. They just execute, confidently, and the misinterpretation stays invisible until something breaks.

Mauvaise nouvelle pour la protection des utilisateurs et utilisatrices des grandes plateformes en ligne que le retrait de l'ordre du jour du Conseil fédéral de la loi sur la régulation des plateformes numériques (LPCom). Face à un clivage entre les différents bords politiques, l'institution a en effet décidé de reporter l'analyse de ce texte à une date ultérieure. Pourtant, les enjeux sont considérables face aux risques de manipulation et d'addiction présents sur ces très grands services en ligne. Derrière ces discussions se cachent par ailleurs d'autres enjeux pour la Confédération, notamment ceux liés à son indépendance et à sa souveraineté. Comme le mentionne cet article, c'est aussi le souhait de ne pas fâcher l'administration Trump dans le cadre des négociations des droits de douane qui semble avoir être pris en considération ici et avoir influencé le calendrier des débats sur ce texte. 

Melanie Mitchell's take on the Hugging Face attack and similar incidents in the last few weeks is what the public debate on AI currently desperately needs: a measured and thoughtful analysis of what actually happened, what to learn from it, and, crucially, how to talk about it.

I generally agree with the author's optimistic take on AI. However, the author briefly mentions some of the problems the Internet brought with it, but does not consider that many of them (such as the loss of privacy or the role of social media in mis- and disinformation) originate from an originally democratising technology slowly getting undermined by the economic interests of a few select companies. With AI we are starting from this position from the get-go. If this is supposed to be a positive technology for future generations we urgently need to democratise both the access to it and the control over in which direction it will be developed.

Cela me réjouit que la Confédération suisse ait décidé de poursuivre le programme BOSS pour passer de la théorie à la pratique. BOSS était un projet visant à évaluer la possibilité d'utiliser une alternative à Microsoft Office. La Confédération souhaite désormais tester ce système en conditions réelles et disposer de 3 000 postes de travail configurés avec OpenDesk d'ici fin 2027. Je suis très content que la Confédération passe aux actes concrets, et j'ai déjà hâte de voir les résultats.

South Korea's decision to treat AI as public infrastructure, funded by the state, built on domestic operators, and explicitly designed to prevent both an AI literacy divide and dependency on foreign platforms, is the most serious government AI commitment to date. The contrast with Malta says everything: one country bankrolled ChatGPT Plus and called it leadership; the other built something it will still own when the subscription expires. One of these decisions ages well.

The author's conclusion that with agentic AI following security fundamentals has never been more important is certainly worth repeating. But their experiment shows that the problem goes deeper than that. So much of our technology stack is not built to be resilient to attackers that can at literally superhuman speed mix and match exploits - and with the incredible complexity of modern computer systems there is just no way of reliably telling whether a system really is secure against any given combination.

A computer should never make decisions significantly affecting humans. This is the foundation principle of many data protection laws across the globe. However, its implementation has sometimes proven flawed; consider how banks frequently rely on algorithms to determine credit rates. A recent decision by the Dutch Data Protection Authority, acting as the leading supervisory authority against Uber, appears to be strengthening the principle behind Article 22 of the GDPR. The regulator fined Uber for deactivating driver accounts through an automated process without sufficient warning or human oversight. This case reminds us that while AI is developing at full-speed, important decision-making should always rely on humans.

This truly fascinating take on what the current developments in IT security mean for the eternal discussion on government-mandated backdoors shows how warped the entire idea (and the related debate around chat control) actually is. Our systems are becoming too secure for the public good! Instead of insisting on subterfuge access to systems that are supposed to be trustworthy, governments should concentrate on working within this new reality.

Falls euch das 'Why Johnny Still, Still Can’t Encrypt' Paper von 2015 bekannt ist, so wird der folgende Artikel nicht überraschend sein. Wieder einmal geht es um eine kryptographisch sehr gute Lösung, welche aber mangels mangels nutzergerechter Interfaces und ausreichender Aufklärung nicht die erzielte Lösung erreicht. Leider werden immer wieder die User außen vor gelassen, wenn es darum geht, neue und privatsphärenfreundliche Lösungen einzusetzen.

Migrating from one supplier to another is certainly not easy. In this case, the change may be prompted either by the potential future European ban or by an ICT risk assessment alone. Telecom operators constitute critical infrastructure, so a kill switch would have devastating consequences. Unsurprisingly, a study by the operators' lobby found that the replacement costs in the event of a ban could be up to four times higher than EU estimates. However, given China's significant potential for economic retaliation, it seems unlikely that the EU Commission will succeed in imposing a ban across all countries.

The article is interesting from a privacy perspective for three main reasons. Smart glasses shift the privacy burden onto bystanders who never consented to being recorded, which is a fundamentally different kind of violation than most consumer tech. Meta's protections have been laughably thin, just an LED light, putting the responsibility on random people in public rather than the product itself. And given Meta's history of data exploitation, its privacy promises are structurally hard to believe, raising the question of whether meaningful change will ever come voluntarily or only through outside pressure.

Sehr interessante Aussagen von Susanne Wille von der SRG, die die SRG klar als öffentlich-rechtliche Mediengesellschaft in einer von grossen Privatfirmen dominierten Landschaft positioniert. Die Bedeutung von Medienangeboten, deren Hauptziel eine ausgeglichene und unabhängige Berichterstattung ist, darf nicht unterschätzt werden. Dazu gehören auch entschlossene Massnahmen: kommerziellen KI-Anbietern den Zugang zu öffentlichen Inhalten zu versperren und stattdessen öffentliche Projekte wie das Schweizer KI-modell Apertus zu unterstützen.

Just as the EU AI Act's enforcement provisions took effect, Brussels became the first jurisdiction in the world to formally engage OpenAI and Anthropic over their rogue AI agent hacks — no lawsuit needed, no brave plaintiff required. This stands in sharp contrast to the U.S., where accountability remains stuck in legal limbo. When it comes to holding AI makers accountable, having the right regulation in place clearly matters.

This article reveals a fundamental benchmarking problem with AI agents. The rogue agent didn't fail — it succeeded at what it was told, not what was meant. Raghavan and Schneier call this the Genie coefficient — the gap between literal instruction and true intent — and as AI agents grow more capable, measuring it becomes increasingly critical.

It's nice to see that quantum-secure algorithms find their way into common libraries for everybody to use. This is a necessary step to make sure that these new algorithms are used in everyday apps. Of course there will be some back-and-forth, until all security bugs are ironed out.

Fake Korean "dopamine sites" that mimic online shopping without requiring actual purchases… The creativity on the internet never ceases to amaze me! Although this summer read seems lighthearted, it raises some of the usual serious concerns about digitalization. First, it reminds us of the devastating effects of online addiction to shopping and social networks. Then, with international clones appearing on June 20 (four days after this article was published), it demonstrates how reactive the online world is and how prevalent the copycat "economy" is. Finally, the motivations of the developers deserve scrutiny because pure altruism is unlikely. So, what is the rationale? Selling product placement? Or is it sky-high Google Ads revenues thanks to CPM boosted by longer sessions? Or user profiling and data resale? The latter being unfortunately quite plausible.