☁️ Cloud sovereignty needs a stronger CADA - Newsletter #60

This may seem like a provocative question to anyone who has been following the discussion on the EU’s digital sovereignty. After all, isn’t the Cloud and AI Development Act (CADA) and the Tech Sovereignty Package supposed to address the EU’s tech dependency and, as the name suggests, increase our tech sovereignty? More specifically, isn’t the package intended to use public procurement to reduce reliance on a small number of non-EU cloud service providers and open up the cloud market?

In a new report, Zuza Warso and Dr Kristina Irion (Associate Professor at the Institute for Information Law (IViR) at the University of Amsterdam) take a close look at Title IV of the CADA proposal, entitled ‘Autonomy’. They analyse the extent to which the measures it proposes can achieve the ambitious goal announced in its title. They identify several issues, the most significant of which is the public order threshold, which would result in around 70 per cent of public sector cloud use being classified at the lowest Union assurance level, with only a minimal level of requirements. This could even prevent public authorities from demanding higher levels of assurance from cloud providers. In light of these issues, they formulate concrete recommendations on how to amend CADA in the ongoing legislative process so that its measures can become a useful tool in Europe’s efforts towards greater tech sovereignty.

One opt-out from AI training for the major AI labs

On 30 October, Anthropic, Google, Microsoft and OpenAI announced that they will honour a shared signal that lets website owners opt out of having their content used to train generative AI models. Instead of writing separate rules for a growing list of AI crawlers, sites will be able to express a single preference in their robots.txt file that all four companies say they will respect, once implemented in the coming months. Others are likely to follow.

The announcement builds directly on work we have been involved in for years. The signal uses the definition of AI training developed in the IETF AI Preferences Working Group, where Paul Keller is one of the editors of the vocabulary, which started from a draft Open Future proposed in March 2025. The overall approach, a shared vocabulary expressed through robots.txt as a first step, is the one we set out in our 2024 paper on implementing rightholder opt-outs.

For now, the signal only covers training, not the inference time uses that rightholders increasingly care about, such as content being used to generate answers in AI search and chatbots. And the companies are moving ahead of the formal standardisation process, as the IETF working group has not yet adopted the standard for the signal. Still, after more than three years of discussion, a working, vendor-neutral opt-out from AI training is a real step forward.

Coming up soon

Digital Commons Policy Summit 2026: As sovereignty, competitiveness and AI dominate the European policy agenda, can the Digital Commons agenda help shape what comes next? Join us on 12 November in Brussels at this yearly event devoted to Digital Commons policies, organised by the NGI Commons Initiative. Open Future will be leading a session on the Digital Commons Strategic Agenda. See more details and register for in-person or online participation here.

Consultation: The European Commission’s targeted consultation on a potential initiative for a better copyright environment for European creativity and innovation closes on 10 November. For inspiration on responding in a way that supports access to knowledge and a sustainable information ecosystem, take a look at COMMUNIA’s preliminary response.

Worth checking out

Interview: Are you interested in AI and the politics of open infrastructures? Katja Mayer and Alek Tarkowski discuss the topic, covering everything from how generative AI intensifies the paradox of open to public AI and new ways of governing digital infrastructure to ensure long-term sustainability.

Commons-based art: Every day, Olof Broström’s Group Photos blends ten openly licensed nature photographs into one image, retaining the focus of each photographer — a demonstration of what the commons make possible.

Article: During Brazil’s presidential election, João Alexandre Peschanski, Amalia Toledo, and Elise Silva argued that the country’s platform rules mistakenly treat Wikipedia like Big Tech, and showed how to regulate the latter without undermining the former.