Market insight

Regulation turns trust into an operating requirement.

In the European Union, transparency in interactions with AI is now a legal obligation. Regulated sectors expect controls, documentation and clear disclosure that a customer is speaking with an AI system.

Key points

Since 2 August 2026 the transparency obligations of Article 50 of the EU AI Act have applied, including the duty to inform people that they are interacting with an AI system.

Obligations for high-risk systems listed in Annex III are now expected from 2 December 2027; whether a contact process falls within them depends on its intended purpose, not on the fact that it is automated.

Consumers want to know when AI is used, and AI adoption in regulated sectors is running ahead of mature governance.

EU AI Act: what applies and when

  • The transparency obligations of Article 50 of the EU AI Act apply from 2 August 2026; what they require depends on the specific use.

    European Commission, Transparency obligations under Article 50 – FAQ, and AI Act Service Desk, Article 50. Regulation; Commission information, not legal advice or a compliance certificate.

  • The Commission published guidelines on the transparency obligations of providers and deployers of AI systems on 20 July 2026.

    European Commission, Guidelines on transparency obligations for providers and deployers of AI systems, 20 July 2026. Interpretative guidance.

  • Earlier stages already apply: the prohibited practices and the AI-literacy requirement (Article 4) from 2 February 2025, and governance and general-purpose AI provisions from 2 August 2025.

    Regulation (EU) 2024/1689 (AI Act), staged application; legal commentary by Sidley Austin, 24 June 2026. Summary of the legal text; not legal advice.

  • Under the Digital Omnibus amendment (political agreement 7 May 2026), obligations for Annex III high-risk systems move to 2 December 2027 and for Annex I systems to 2 August 2028; Article 50 was not deferred.

    Legal commentary: Lexology (1 July 2026), Pinsent Masons (6 July 2026), Lewis Silkin (9 July 2026). Secondary sources recorded in September 2026; dates should be checked against the consolidated text in EUR-Lex.

  • Systems placed on the market before 2 August 2026 have until 2 December 2026 to meet Article 50(2) on the marking of synthetic content.

    Legal commentary by Sidley Austin, 24 June 2026. Secondary source.

  • AI systems used to evaluate the creditworthiness of natural persons are among the high-risk uses listed in Annex III; high-risk classification follows a system’s intended purpose rather than the fact that a conversation is automated.

    Regulation (EU) 2024/1689, Article 6 and Annex III. Summary of the legal framework; not legal advice.

Trust and disclosure

  • 56% of consumers say knowing that AI is being used matters more than simply getting the fastest result; 71% are concerned that an AI agent acting on their behalf could misrepresent their personality or values.

    Ping Identity, 2026 Global Consumer Survey: AI Trust (Talker Research). Vendor-sponsored; 11,000 consumers in 12 countries; agentic AI generally, not telephony alone.

  • Only 37% of CX leaders said they provide transparency about their use of AI, while 95% of consumers want explanations.

    Zendesk, CX Trends 2026. Vendor research.

  • 58% of consumers were only somewhat or not at all comfortable using AI tools to engage with brands; 29% had stopped using a brand because of poor customer experience.

    PwC, 2025 Customer Experience Survey: The Loyalty Illusion, 29 September 2025. US; 5,511 consumers and 406 executives; covers all channels.

  • 86% of consumers were willing to share data with companies that are transparent about how it is used.

    Qualtrics, 2026 Consumer Experience Trends Report, October 2025. 20,000 consumers in 14 countries; multichannel.

Governance in regulated sectors

  • 92% of 180 German telecoms, media and technology companies use AI, but only about 10% have a mature AI strategy; 79% expect benefits in customer service.

    KPMG Germany, TMT Monitor 2026: Adoption grows faster than governance, 21 May 2026. Germany; TMT sector.

  • The UK telecoms regulator published a discussion paper examining customer preferences, transparency and the risks arising from the use of AI in telecoms customer service.

    Ofcom, AI in telecoms: impact on customers – discussion paper, 25 September 2026. United Kingdom; Ofcom rules are not the same as EU obligations.

Transparency as an operating requirement

Vee's interpretation. The analysis below is our own reading of the sources above, not a finding of the organisations cited.

For banks, insurers, telecoms operators and healthcare providers, an AI contact process now has to disclose itself, and the institution has to be able to show that it did. Article 50 duties can fall on the provider, the deployer or both depending on role and design, so contracts need to state who implements the disclosure and who keeps the evidence.

Disclosure interacts with the customer experience: how and when the customer is told affects whether the conversation continues. That is a design and testing question as much as a legal one.

Where the high-risk line falls

Automating a conversation, even at large scale, does not by itself make a system high-risk. The heavier regime is triggered by intended purpose – for example, using AI to assess a person’s economic situation or to determine access to a service.

For contact processes in consumer finance and collections, the practical boundary lies between delivering information and recording a customer’s declaration on the one hand, and using AI to decide or recommend materially different consequences for an individual on the other. Where the second applies, documentation, human oversight and the ability to reconstruct what happened become substantially more demanding.

Architecture matters for cost. Systems that generate content freely, adapt after deployment or cannot be reconstructed call for heavier monitoring and evidence management than systems built on approved scenarios and logged transitions.

By sector

Banking and consumer finance carry the clearest Annex III exposure, through the assessment of creditworthiness. In telecommunications, adoption is broad but, on KPMG’s German evidence, governance is less mature; in the United Kingdom, Ofcom has opened its own discussion of AI in customer service. Healthcare contact involves sensitive data and callers who may be elderly, unwell or under stress, which raises the bar for disclosure, escalation and the reconstruction of what was said.

The research we hold does not analyse sector-specific AI Act exposure for insurance or for energy and utilities; for those sectors the general framework above applies, and the classification of any given process depends on its intended purpose.

What to watch

2 December 2026: end of the transition period for Article 50(2) for systems placed on the market before 2 August 2026, according to legal commentary.

2 December 2027: expected start of obligations for Annex III high-risk AI systems under the amended timeline.

2 August 2028: expected start of obligations for high-risk systems covered by Annex I (products), under the same amendment.

Supervisory practice on Article 50 disclosure by national authorities and further Commission guidance during 2027.

Vee in context

Vee’s consolidated annual report for 2025 (management report, p. 25) states that the growing importance of AI regulation, including the EU AI Act, increases the importance of solutions that are auditable, controllable and reconstructable at the level of the decision process, and that Vee S.A. intends to use this as an element of competitive advantage, particularly in regulated sectors. The same report (pp. 27–28) describes a range of Brilliance architectures, from the fully deterministic Brilliance H to the more generative Brilliance 4, and the Management Board’s assessment that interest in fully generative solutions for critical processes in regulated European sectors is significantly lower than in deterministic or hybrid architectures. The Q3 2026 report (letter, p. 12) states that control, supervision and security mechanisms are being developed alongside wider automation.

Market data on this page come from third-party publications and are attributed to their authors. They describe the market, not Vee’s results. Forecasts are the publishers’ own and are not forecasts by Vee S.A. Figures were checked against the publishers’ pages on the date shown.

Other market insights

The AI value gap

Spending on AI in customer service is rising, but few organisations report the return they expected. The gap sits between deploying a model and running a process that delivers a measurable result.

Read the insight

Integration and legacy

In large organisations, AI meets existing telephony, CRM, collection and policy systems. Most barriers reported by enterprises are about connecting to those systems safely.

Read the insight

The human route

Automation is accepted when it solves the issue and when a person is reachable if it does not. Enterprises are redesigning human roles rather than removing them.

Read the insight

Contact data

Organisations record almost every call, yet analyse only a small fraction. Process data is where measurable improvement comes from.

Read the insight

Service economics

Budgets are moving towards AI, but the cost of automated resolution is not falling to zero. Buyers increasingly ask what a process achieves, not how many minutes it uses.

Read the insight