National models and the world's first AI minister
Updated: Nov 24, 2025
AI Business Risk Weekly
This week saw some firsts: Switzerland launched an open-source national AI model that could set a new standard for transparency, Albania put an AI system in charge of public procurement, and Europe’s new Data Act quietly took effect with sweeping rules on data sharing. Meanwhile, Google released a model trained with built-in privacy protections, and fresh usage studies showed just how much personal AI use has outpaced workplace adoption. We round out with a few free ways to help your team keep up on governance basics.
Switzerland bets on a national model. Will other countries follow?
Switzerland’s top research institutes released Apertus, a multilingual AI model that is not only open-source but fully documented from datasets to intermediate checkpoints. The model was designed with legal compliance in mind, respecting copyright opt-outs and EU-style privacy rules. Beyond the technology, Apertus is a statement: a country showing it can build its own system, aligned with its own values and languages. Some observers are already asking whether this could spark a “Bern Effect,” where other European nations launch national models of their own.
Business Risk Perspective: Buyers may start asking not just “does this model work,” but “whose norms does it reflect?” For companies, that means the country of origin could become as important as the benchmark scores.
Albania names the world’s first “AI minister”
Albania’s Prime Minister announced Diella, an AI system that will oversee all public procurement decisions, with the goal of eliminating corruption. Represented by an avatar dressed in traditional Albanian clothing, Diella builds on Albania’s e-government platform but now carries real decision-making power over tenders, raising questions about bias, contestability, and what due process looks like when an algorithm sits in the minister’s chair.
Business Risk Perspective: This sets a surprising precedent: businesses may soon find themselves “pitching” to LLMs instead of people. Whether that means a more level playing field or new forms of hidden bias is still an open question.
The EU Data Act comes into force
On September 12, 2025, the EU Data Act officially became applicable. The law requires that data generated by connected products and services, including everything from smart appliances to industrial equipment, can be shared fairly with users and third parties. It also bans “vendor lock-in” practices by forcing cloud providers to make switching easier. The goal: give businesses and consumers more control over the data their devices produce.
Business Risk Perspective: Compliance now means proving that customers can access, share, and even migrate their data without hitting artificial barriers.
Google’s VaultGemma brings privacy into the training pipeline
Google unveiled VaultGemma, a new AI model trained with differential privacy, a method that adds statistical “noise” during training so the system doesn’t memorize individual data points. Alongside the release, Google published new research that lays out the trade-offs between accuracy, compute costs, and privacy guarantees. The takeaway: it’s now possible to train large models that don’t regurgitate personal or sensitive information, even if the output isn’t cutting-edge.
Business Risk Perspective: Models trained this way promise a real reduction in leakage risk. Over time, “we trained it private” could become a trust badge customers look for as much as accuracy.
How we actually use AI at work
Fresh usage studies from OpenAI and Anthropic highlight how habits are shifting. ChatGPT is used most for writing and decision support, while Claude leans toward coding and analysis. Personal use is exploding. ChatGPT’s non-work use grew from 53% of messages in 2024 to 73% this year. Adoption is uneven across regions too, with Claude concentrated in wealthy countries and ChatGPT growing fastest in emerging markets.
Business Risk Perspective: Culture is moving faster than policy. If most use is happening outside official channels, companies may inherit the risks of personal AI adoption whether they plan for it or not.
And finally, three free ways to get your team fluent in AI governance
If you want a low-stakes way to bring your team up to speed, here are three free certifications worth checking out:
AIQI’s ISO/IEC 42001 overview — a primer on the new international standard for AI management systems.
Alison’s AI Governance and Ethics — a course covering the ethical side of AI deployment.
Microsoft Learn: Responsible AI — a hands-on look at how one of the biggest players frames responsible AI.
Business Risk Perspective: AI governance is the missing piece in many AI rollouts. Even without official rollouts, most professionals are using AI tools with no formal training. These courses give teams a starting point to ask the right questions, share a common language, and avoid costly mistakes before they happen.
AI Business Risk Weekly is a Conformance AI publication.
Conformance AI ensures your AI deployments remain safe, trustworthy, and aligned with your organizational values.
