Vendor AI: Who is responsible for AI you did not buy as AI?

What is vendor AI?

Vendor AI is AI inside third-party systems that can trigger deployer duties without any in-house model development. When financial institutions think about artificial intelligence, they usually think of conscious decisions: buying a new AI tool, starting a pilot project, approving a budget. Reality is often different. AI may already be part of the existing software landscape.

This is vendor AI: AI functionality in systems from external providers. For banks and insurers it is critical because deployer duties may arise even if the institution did not develop the model itself.

Which systems typically contain vendor AI?

Six areas from core banking to chatbots; credit decisions may touch Annex III No. 5(b), while fraud detection is excluded.

The providers named are illustrative examples of typical product categories. Whether and which AI functions are active in the product version actually deployed can only be determined from the manufacturer's product documentation; a provider name alone implies neither AI use nor a high-risk classification. ICT risks from purchased AI services are additionally governed by DORA third-party management; the joint EBA/EIOPA/ESMA statement of 31 July 2026 on frontier AI models is current supervisory guidance on this.

Role shift under Art. 25: When does a deployer become a provider?

Art. 25(1) EU AI Act determines when a deployer, distributor or importer is considered the provider of a high-risk AI system and takes over the provider obligations. It covers three cases, each tied to high-risk:

• Name or trademark: the institution puts its own name or trademark on a high-risk AI system already placed on the market or put into service.
• Substantial modification: the institution substantially modifies a high-risk AI system in a way that it remains high-risk.
• Change of intended purpose with high-risk effect: the institution modifies the intended purpose of an AI system (including a GPAI system) so that it becomes high-risk.

Merely using a purchased application within the institution is not a rebranding of the AI system and does not trigger this role change. This is distinct from the general provider definition in Art. 3(3) (developing or having developed and placing on the market under one's own name) and from the separate role of the GPAI model provider. The assessment depends on the individual case. A wrong assumption that the institution is only a deployer can create significant gaps in documentation, conformity and supplier control.

CSRD / ESRS: Is vendor AI part of the value chain?

Yes, vendor AI can also be relevant for CSRD / ESRS. CSRD requires reporting on own operations and, where applicable, value chains, products, services and business relationships. If purchased AI systems cause discrimination risks or human-rights impacts, procurement, supplier management and governance need to be prepared. This presupposes that the institution is subject to reporting at all: Directive (EU) 2026/470 has limited the CSRD scope to EU companies with more than 1,000 employees and more than EUR 450 million net turnover; national transposition, transition years and group-level assessment must also be checked. The impact or risk must also be material – not every bias incident leads to an ESRS disclosure.