Skip to content

Democratizing large-scale AI for the benefit of society: Open calls for disruptive ideas

In addition to its core research activities as outlined below, the Swiss AI Initiative is distributing 10-20 million GPU hours in 2025 for disruptive research projects through open calls. We look for research projects that aim to contribute to advances in AI fundamentals or impactful applications of AI. Researchers outside of Switzerland are encouraged to apply if they team up with at least one of our PIs and aim to create novel open science artifacts that benefit the Swiss, European or global ecosystem and societal context.

Reviewing the Cryptography Used by Signal

Soatok is an opinionated cryptographer, but a very good one. After having looked and complained about many ‘private messenger apps’, he has a grab at Signal. But he leads us through his non-findings in a very nice way, which also shows how Signal’s cryptography is chosen and made secure. A very intense and technical read (…)

Disappearing Digital Books: Amazon Reminds Us Why Bookshelves Beat the Cloud

With physical media, no one can reach into your home and remove what you bought from your shelf.’ Not only can Amazon delete books remotely, the shift towards cloud-based storage for e-books means that copies can be modified by Amazon and publishers at will. With the rapid pace of change in the political environment, it (…)

SF-GWAS, un outil pour s’entraîner sur des génomes en toute sécurité

L’EPFL, en collaboration avec le MIT et Yale, a mis au point, SF-GWAS, un outil pour entraîner des intelligences artificielles sur le génome humain, sans partager les données. SF-GWAS est déjà utilisé dans les cinq hôpitaux universitaires suisses.
Les explications de Jean-Pierre Hubaux, directeur académique du Center for Digital Trust de l’EPFL, au micro de Stéphane Délétroz.

Navigating Trust in Transformative Technologies

The text explores the concept of trust in transformative technologies, highlighting the importance of understanding the nuances of “trust” in different contexts. It elevates discussions to a philosophical level by questioning the true essence of trust—whether it is rooted in transparency, reliability, or alignment with human values. By addressing these contrasts, the article encourages dialogue (…)

AI systems with ‘unacceptable risk’ are now banned in the EU

The European Union’s AI Act has reached a significant milestone by banning AI systems across the EU that pose an ‘unacceptable risk,’ defined as serious threats to people’s rights and safety. The Act will be implemented in phases, with additional regulations for high-risk AI systems expected in the next 2-3 years. With security risks from (…)

Google Lifts a Ban on Using Its AI for Weapons and Surveillance

Following in the footsteps of OpenAI, Anthropic and Meta, Google is the latest Big Tech giant to lift restrictions on their AI technologies, in this case with regards to, amongst others, weapons and surveillance. These companies reassuring us over and over again that they were committed to safe and ethical use of AI was a (…)

US immigration is gaming Google to create a mirage of mass deportations

A very dark example of search engine optimization (SEO): Ice is suspected of updating time stamps of reports on old operations to push them up in Google searches. Whether it is, as the article suggests, done on purpose to create Potemkin-like images of mass deportations, or was just a clumsy update of their website, this (…)

A look back on matrix.epfl.ch in 2024: A Success Story!

We’re thrilled to announce the publication of our latest blog post exploring the remarkable year 2024 of matrix.epfl.ch. 📚In 2024, the Center for Digital Trust discovered matrix.epfl.ch hosted by Edwand Andò from EPFL’s Center for Imaging and decided to offer our support. This led to a fruitful collaboration in 2024 to maintain and develop this (…)

Matrix at EPFL

Matrix is a widely used open source chat platform. It is an alternative to platforms such as Slack and MSTeams. In line with C4DT’s objective to promote services in the realm of digital trust, we set up our own Matrix instance and migrated our professional communications there in 2019. A privacy-first approach with encrypted chat (…)

The biggest breach of US government data is under way

Amid the controversy surrounding Musk’s operatives accessing US government data, this article gives some nice examples of how an organization’s sensitive data could become compromised. At the top of the list is the failure to follow basic cybersecurity protocols, such as adhering to transparently developed guidelines for system access and data handling. Other red flags (…)

Ciné-débat : Quand la cybersurveillance menace les droits humains

Entre l’espionnage de nos conversations WhatsApp et l’activation à distance du micro ou de la caméra de notre téléphone, le film “Cybersurveillance, un impact planétaire” révèle comment des gouvernements utilisent des technologies de surveillance pour cibler certains de leurs citoyens et interroge notre rapport aux libertés fondamentales.

DOGEs and don’ts

Donald Trump’s new Department of Government Efficiency, led by Elon Musk, is set to transform the US Digital Service into the US DOGE Service, ostensibly to modernize federal technology. However, there are indications that the true focus is on cost-cutting and workforce reduction, as evidenced by executive orders mandating a government-wide hiring freeze and allowing (…)

The Less People Know About AI, the More They Like It

Do you believe in magic? Recent research reveals a ‘lower literacy-higher receptivity’ link, suggesting that the less people understand AI, the more they perceive its human-like capabilities as magical, making them more likely to use it. This quirky perspective on how people perceive AI could reveal new ways to communicate its benefits and risks to (…)

AI-generated child sex abuse images targeted with new laws

The new laws targeting AI-generated CSAM position the UK as the first nation to criminalize the possession, creation, or distribution of AI tools used for generating such material, imposing penalties of up to five years in prison. These measures also prohibit related manuals, penalize those operating websites that share CSAM, and grant Border Force the (…)

Meta, X sign up to Euro Commish code of conduct on hate speech

While the ‘Code of conduct on countering illegal hate speech online’ that the European Commission included into the Digital Services Act (DSA) is work in progress, the fact that even companies such as Meta and X feel compelled to sign shows that regulations are far from the toothless tigers that they are often made out (…)

The Trump Cryptonaissance Has Arrived

Cryptocurrencies are in again thanks to Donald Trump. Is the U-turn the new government is taking on digital assets due to careful deliberations of the advantages and risks? Is its goal to take currencies into the 21st century and beyond? Or is it nothing but the result of an extremely successful lobbying campaign? This may (…)

US Cloud soon illegal? Trump punches first hole in EU-US Data Deal

We’ve been having a lot of discussion about the ‘Swiss Government Cloud,’ which is effectively hosted in the USA and China. The EU has the same setup, but someone – likely in the Trump administration – is now attempting to put holes in it by removing parts of the privacy guarantees its remote cloud has. (…)