AI evidence breaks at organisational boundaries.
A shared approach to carry and verify AI evidence across organisations. Five European organisations are joining forces to change that.
- decision
- loan application declined
- model
- credit scoring, version used
- data
- training data, provenance
- checked by
- accountable actor
- rule
- applicable obligation
Each piece of evidence exists. Put together, it rarely adds up to proof anyone can verify.
The story is scattered
When a bank turns down a loan with the help of AI, or a hospital relies on a model to read a scan, someone must later be able to explain what happened: which model was used, trained on which data, checked by whom, under which rule. Today that story is scattered between the model's maker, the organisation that uses it and the auditor.
The rule is coming
From December 2027 (August 2028 for AI built into products such as medical devices), the AI Act requires high-risk AI systems to keep records of how they operate. Those records only count if others can read and trust them.
Data sharing is held back
Whoever comes next in the value chain must be able to know that checks were done, without repeating them. Without that, exchanges of data and models, for example in European data spaces, stay limited to a few trusted partners and do not scale.
A common specification, written by the members
The Alliance will define a common specification to produce and verify evidence on AI across organisations, without anyone handing over models or data: organisations exchange verifiable attestations instead. Members will work towards simpler, more efficient components that integrate easily into the systems organisations already run.
Dual-use
Civil and defence applications.
Finance
Financial services.
Pharma and health
The pharmaceutical industry and health.
Other sectors
Other sectors will follow.
Five European organisations that already work together
They cooperate in EU research projects and in European data-space initiatives. The Alliance gives that cooperation a shared goal.
Why the founding members are joining
“Today, the evidence behind an AI decision breaks as soon as it leaves the organisation that produced it. Cloud Data Engine already automates the verification of regulatory compliance across federated ecosystems and has built a decentralised registry of qualified AI models. We bring that experience to the Alliance so that evidence on AI stays continuous from the model's maker to the auditor, with components light enough to plug into the systems organisations already run.”
“To utilize the full potential of AI, we need to establish evidence and trust across actors. AIT brings its research on secure and trustworthy digital systems so that each link of the chain can rely on the checks made before it.”
“Data spaces have shown that organisations can share data while keeping control of it. The Alliance applies that principle to evidence about AI, so that it can be shared, checked and trusted along the whole chain.”
“Our clients in defence, public sector, financial services, health and industry ask us how to demonstrate that the AI they use, and the AI their suppliers use, is under control. The Alliance turns that demand into one answer that works across organisations.”
“TECNALIA is working to ensure the traceability of compliance evidence across the full lifecycle of high-risk AI systems and AI embedded in products. The approach addresses two fronts: first, establishing robust evidence-based traceability frameworks in the health and pharmaceutical sectors, where regulatory requirements are particularly demanding; second, providing structured accompaniment to manufacturers of high-risk AI systems under Annex I, with a particular focus on the energy sector. In both cases, the objective is the same: delivering traceability components simple enough for industrial users to adopt and integrate into their existing compliance workflows.”
We start in Q1 2027
Launch in Vienna
Announced in the presence of all founding members.
Governance set up
Membership rules defined. Members commit to contribute.
First deliverables
First technical deliverables and pilots with industrial users.
Press release and logo
Download the launch press release in the language of your choice, and the Alliance logo.
Press contact: press@evidencealliance.ai (Jean-Marie Davesnes)
Register your interest
We first want to hear from organisations that use AI in defence and dual-use, finance, pharma and health, and from those that audit or supervise them. Membership rules will be set in early 2027, and members commit to contribute.
Journalists: choose "Press or media" in the form, or write to press@evidencealliance.ai.