Εθνική Στρατηγική για την Τεχνητή Νοημοσύνη (ΤΝ) της Κυπριακής Δημοκρατίας

SUBMISSION TO THE PUBLIC CONSULTATION ON THE NATIONAL AI STRATEGY OF THE REPUBLIC OF CYPRUS 2032

Recognising Multimodal and Embodied Conversational Interfaces as a National Accessibility Capability

Submitted by: Cyber Leo Limited (trading as RAVATAR), HE384983, Limassol, Cyprus
Contact: Ruslan Synytskyy, Chief Executive Officer
Email: rs@ravatar.com
Date: 29 July 2026

EXECUTIVE RECOMMENDATION

The Strategy correctly identifies conversational AI and multilingual assistants as reusable national services rather than isolated sector applications. We support this framing without reservation.

We recommend one clarification, applied consistently across the Strategy: that conversational AI be specified as multimodal, capable of speech, visual presence and natural interaction, and not implicitly limited to text-based interfaces.

This is not a technical preference. The Strategy commits Cyprus to digital inclusion for people with disabilities and for pensioners, and to multilingual access. A text-only interface systematically excludes the populations the Strategy names as priorities: older citizens, people with low digital literacy, people with visual or motor impairments, and non-Greek-speaking residents. Where the interface is the service, the modality of the interface is an accessibility decision, not an implementation detail.

We also recommend that the Strategy establish a national transparency standard for synthetic-persona interfaces, so that Cyprus sets the governance rules for this technology rather than importing them.

The six comments below address:

1. Specification of shared conversational AI services as multimodal
2. A national disclosure standard for synthetic-persona interfaces
3. Accessibility requirements for the Virtual AI Patient Orchestrator
4. Scope of the tourism SME co-funding scheme
5. Sovereign deployment of the citizen-facing interface layer
6. A glossary definition for multimodal conversational interfaces

All proposed amendments are drafted in capability terms and are technology-neutral. None names or favours a specific product, vendor or architecture.

COMMENT 1: SPECIFY SHARED CONVERSATIONAL AI SERVICES AS MULTIMODAL

SECTION REFERENCE

Section 3.4.4 (Shared Capabilities and Reuse); Annex C, Government and Public Sector, Implementation, evidence and compliance; Annex B, Control Gate 1, category “reusable common services”.

COMMENT / SUGGESTION

The Strategy states that Cyprus will build shared, reusable AI services including “conversational AI for continuous citizen service delivery” and “multilingual citizen assistants”, exposed through “a common, modular intelligent government platform”.

We suggest these references be amended to read, in substance:

“conversational AI, including multimodal and embodied interfaces supporting speech, visual presence and accessible interaction, for continuous citizen service delivery”

and that the shared services catalogue explicitly list a multimodal interaction layer as a reusable component alongside identity, payments, document intelligence and case management.

JUSTIFICATION

The Strategy commits, in Section 3.9 and in the OECD-aligned principles at Annex A, that “new solutions must cater to multilingual support and inclusive access tools that ensure that people with disabilities can interact with the services”, and in the Executive Summary to promoting digital inclusion “including people with disabilities and pensioners”.

Section 3.4.4 establishes that capabilities built once will be reused across government. This makes the specification of the shared conversational service unusually consequential: whatever modality is scoped into the reference architecture will propagate into every ministry that adopts it, and will be difficult to retrofit once the platform is procured. The Strategy notes that procurement for the first six transformational solutions is intended to launch within eight months. The specification decision is therefore imminent.

The current wording does not exclude multimodal interfaces. It also does not require them, and in procurement practice an unspecified requirement is an absent requirement. Given that the Strategy sets an explicit target of reducing citizen waiting times by 40% and improving citizen experience, and given that the citizens least well served by existing digital channels are precisely those least able to use a text interface, we consider the clarification proportionate and low-cost.

A second consideration: architectural optionality over a seven-year horizon.

Section 3.7.3 provides that infrastructure choices “should therefore be guided by considerations of interoperability, portability, vendor diversity, supply-chain resilience, security, and long-term sustainability”, and that this “reduces lock-in risks while preserving strategic flexibility”. Section 3.13.4 commits the Strategy to remaining “adaptive rather than static” in the face of technological change.

We suggest these principles apply with particular force to the interaction layer, and that the Strategy would benefit from stating so explicitly.

The Strategy runs to 2032. The shared conversational service will be specified once and inherited by every ministry that adopts it, across the sixteen transformation projects and beyond. Interaction modality is an architectural property of that service: a platform designed for multimodal interaction can serve text, whereas a platform designed for text cannot later serve speech and visual interaction without substantial rework or re-procurement.

The asymmetry is the point. Requiring modality-extensibility in the initial specification is close to costless, because it constrains architecture rather than mandating deployment. Retrofitting it into a procured national platform is expensive, slow, and may not be practicable within the Strategy’s timeframe. This holds regardless of how interaction preferences evolve over the period, which is why we put it forward as a resilience argument rather than a forecast.

We would specifically suggest that shared conversational services be required to expose a modality-independent interaction layer, so that additional modalities can be added without re-architecting the underlying service, and that this be treated as a portability requirement of the kind Section 3.7.3 already contemplates.

We note that this comment asks for a specification of an existing commitment, not for a new programme or new expenditure.

COMMENT 2: ESTABLISH A NATIONAL DISCLOSURE STANDARD FOR SYNTHETIC-PERSONA INTERFACES

SECTION REFERENCE

Section 3.9.3 (Transparency and Explainability); Section 3.9.6 (AI Social Contract Principles); Annex B, Control Gate 2.

COMMENT / SUGGESTION

We recommend the Strategy commit to a national transparency standard governing AI systems that present a synthetic human persona, whether visual, vocal or both. The standard should require, at minimum:

– Persistent disclosure. Continuous, unambiguous indication that the citizen is interacting with an AI system, present throughout the interaction and not only at its start.

– No impersonation of identifiable individuals. Synthetic personas deployed in public services must not replicate the likeness or voice of a real, identifiable person without that person’s explicit, documented and revocable consent.

– One-step human escalation. A visible, always-available route to a human official, with no requirement to restate the request.

– Provenance and audit. Recorded lineage of the persona asset, the underlying models, and the interaction logs, consistent with the auditability requirements already set out in Section 3.9.3.

– Proportionality. Requirements scaled to risk, consistent with the EU AI Act’s risk-based approach, with the strictest application to health, benefits and legal contexts.

JUSTIFICATION

Article 50 of the EU AI Act establishes transparency obligations for systems that interact directly with natural persons and for systems generating synthetic audio, image, video or text. It requires that this information be provided in a clear and distinguishable manner at the latest at the time of first interaction, and that it conform to applicable accessibility requirements. The Strategy itself notes the 2 August 2026 compliance milestone. Synthetic-persona interfaces are an area where the Act sets a floor and where national implementation guidance does not yet exist in most Member States.

We note that the Act’s own linkage of transparency to accessibility supports the point made in Comment 1: disclosure that a citizen cannot perceive is not disclosure. The Strategy commits to leveraging EU AI Act alignment as a source of competitive advantage rather than treating it as a constraint, and this is an area where that ambition can be made concrete.

Cyprus has an opportunity to publish that guidance first. The Strategy states an ambition for Cyprus to be “a testing and deployment environment for EU and regional markets, particularly for regulated and multilingual contexts”. A clear, published national standard for trustworthy synthetic-persona deployment would be a concrete instance of that positioning, and would give operators regulatory certainty in a technology area where uncertainty is currently the main barrier to adoption.

There is a public trust argument as well. The Strategy is explicit that “AI systems must be explainable, auditable, and understandable to those they affect”, and that citizens interacting with AI must be made aware of it. Embodied interfaces raise this question more sharply than text interfaces do. Addressing it in the Strategy rather than after the first incident is the lower-risk path.

We raise this comment notwithstanding that stricter rules apply to our own product category. We consider a published standard preferable to an unregulated market, both for citizens and for responsible operators.

For transparency as to our own position: we are currently implementing AI disclosure, human oversight and interaction logging as architectural requirements of our platform, together with a formal EU AI Act conformity assessment, under a dedicated compliance work package in our European Innovation Council Accelerator programme application. The measures proposed above are therefore ones we are building to, not ones we are proposing for others.

COMMENT 3: SPECIFY ACCESSIBILITY REQUIREMENTS FOR THE VIRTUAL AI PATIENT ORCHESTRATOR

SECTION REFERENCE

Annex C, Healthcare and Life Sciences, Flagship programme(s).

COMMENT / SUGGESTION

The Strategy describes the Virtual AI Patient Orchestrator as “an intelligent, multilingual Virtual AI Patient Orchestrator that functions as a national digital front door”, operating “under clinician oversight, with safety nets for vulnerable populations”.

We suggest the programme description specify that the front door must be accessible by voice and visual interaction, not by text alone, and that accessibility for older patients and patients with disabilities be included among the programme’s evaluation indicators rather than treated as a downstream implementation concern.

We further suggest that a bounded pilot be conducted before national rollout, comparing interaction modalities on a defined patient cohort, with pre-registered indicators covering completion rate, comprehension, time to first clinical assessment and patient-reported experience, disaggregated by age group and language.

JUSTIFICATION

The Strategy already commits to evaluating national triage and care navigation pilots “against WHO-aligned indicators, focusing on triage accuracy, time to first clinical assessment, and reduction of avoidable visits”. Adding accessibility indicators to that set is consistent with the existing evaluation design and adds little cost.

The clinical case is straightforward. A digital front door that is intended to reduce pressure on emergency services succeeds only if the patients most likely to present at emergency departments can use it. Those patients are disproportionately elderly, and elderly patients in Cyprus include a substantial cohort with limited digital literacy. A front door that only the digitally confident can open will shift load rather than reduce it, and may widen rather than narrow the access gap the programme is meant to close.

The Strategy identifies “AI theatre” as a national risk: many pilots, little measured impact. A pre-registered, modality-comparative pilot with disaggregated outcomes is a direct mitigation of that risk and would produce evidence usable across the other citizen-facing programmes.

COMMENT 4: EXTEND THE TOURISM SME CO-FUNDING SCHEME TO MULTIMODAL CONCIERGES

SECTION REFERENCE

Annex C, Tourism and Hospitality, Implementation (Empowered Digital Tourists programme); and the pillar deliverables “Support for sustainable hotels and SMEs” and “Digital concierges and transparent AI use”.

COMMENT / SUGGESTION

The Strategy provides that the Empowered Digital Tourists programme “co-funds multilingual, privacy-compliant GenAI chatbots and recommendation systems for sustainability-certified hotels and tourism SMEs”, and separately anticipates “GenAI-based concierges supporting multilingual guidance, culture and heritage explanations, and human escalation”.

We suggest the co-funding provision be worded to cover multilingual, privacy-compliant conversational AI systems including chatbots, voice assistants and embodied digital concierges, so that the funding instrument matches the concierge capability the same pillar describes.

JUSTIFICATION

As currently drafted, the deliverables table describes digital concierges while the funding mechanism references chatbots. Aligning the two removes an inconsistency and avoids a situation in which the stated capability is not eligible for the stated instrument.

The substantive case is that hospitality is an in-person, multilingual, service-quality-driven sector, and the Strategy is explicit that AI adoption in tourism must preserve “the hospitality, warmth, and cultural authenticity that distinguish Cyprus”. A text chatbot is a poor expression of that objective. A concierge that speaks, in the visitor’s language, with visible AI disclosure and immediate human escalation, is a closer fit to both the service standard and the transparency requirement.

We note the Strategy’s own cited evidence that 8% of chatbot conversations resulted in a confirmed booking, rising to 25% with a single follow-up, with an estimated 10% increase in direct sales. Those figures are drawn from text-based deployments. Extending eligibility would allow Cyprus to generate comparative national evidence on whether multimodal interfaces improve on that baseline, which is information no other Member State currently holds.

We also support the Strategy’s emphasis that this scheme reach SMEs and local communities and not only large hotel chains. Interface quality is one of the few areas where a small operator can match a large chain if the tooling is accessible, and the co-funding mechanism is well designed for that purpose.

COMMENT 5: TREAT THE CITIZEN-FACING INTERFACE LAYER AS A SOVEREIGN CAPABILITY

SECTION REFERENCE

Section 3.7.3 (Digital Sovereignty and Strategic Autonomy); Section 1.4.6; Executive Summary, “AI sovereignty leveraging partnerships”.

COMMENT / SUGGESTION

We suggest that the Strategy identify the citizen-facing conversational interface layer as among the critical layers over which Cyprus should retain sovereign control, alongside compute, data and skills, and that shared conversational services procured for government be required to support deployment within national infrastructure.

JUSTIFICATION

The Strategy frames the national choice as being between “coordinated, sovereign, human-centred AI on Cyprus’s own terms, or piecemeal adoption of foreign platforms that gradually lock the country into other people’s economic and governance models”. It commits to Cyprus “controlling all critical layers, including sovereign compute, governed access to high-value data, core skills, and national AI assurance”.

The interface layer belongs on that list. It is the layer through which every citizen interaction passes, where the transcript of that interaction is generated, and where the tone, language and cultural register of the state’s relationship with its citizens is set. A conversational interface that can only run as a foreign-hosted service creates a dependency at precisely the point of highest sensitivity, and constrains the data-residency and audit guarantees the Strategy commits to elsewhere.

This is not a hypothetical concern. The established providers of multimodal and avatar-based conversational interfaces are, to our knowledge, almost entirely non-EU: Synthesia (United Kingdom), HeyGen (United States), D-ID (Israel), DeepBrain AI (South Korea) and UneeQ (United States). We are not aware of a major EU-headquartered provider in this category. Where a capability category has no European supply, adopting it without a sovereignty requirement produces exactly the dependency the Strategy warns against, and does so silently, because the dependency is created at procurement rather than announced as a policy choice.

The corollary is that a sovereignty requirement in this layer is also an industrial policy instrument. It creates addressable demand for European and Cypriot suppliers in a category where European capability is currently thin, which is consistent with the Strategy’s stated ambition to move Cyprus “from consumer to creator of AI solutions”.

Requiring deployability within national infrastructure does not require Cyprus to build the technology itself, and is consistent with the Strategy’s partnership model. It requires only that the procurement specification permit it, which is a question of how the requirement is written and is therefore appropriate to settle at strategy stage.

This provision would also align the shared services layer with the G-Cloud programme described in Annex C, which is intended to “provide a fully governed, hybrid cloud foundation, expand the AI compute capacity, and establish governed sandboxes”.

COMMENT 6: ADD A GLOSSARY DEFINITION FOR MULTIMODAL CONVERSATIONAL INTERFACES

SECTION REFERENCE

Annex G (Glossary of Key Terms).

COMMENT / SUGGESTION

We suggest the Glossary include an entry along the following lines:

“Multimodal Conversational Interface. An AI system that interacts with a person through more than one channel, such as speech, visual presence, gesture or text, rather than through text alone. Such systems are commonly referred to as conversational avatars or digital humans. They are used to improve accessibility and comprehension for users with limited digital literacy, visual or motor impairments, or limited proficiency in the interface language, and are subject to the transparency obligations applicable to AI systems that interact directly with natural persons.”

JUSTIFICATION

The Strategy refers to conversational AI, multilingual assistants and digital concierges in several places without defining what interaction modalities those terms encompass. The Glossary states that its purpose is to “support consistent understanding and interpretation of the Strategy”. A definition would resolve an ambiguity that will otherwise be resolved differently by each implementing body.

Including the common commercial terminology in the definition, while keeping the operative term capability-based, allows the Strategy to remain technology-neutral in its requirements while ensuring that implementers, procurement officers and suppliers interpret those requirements consistently. Terminology in this field is not yet settled, and a national definition reduces the risk of divergent interpretation across the sixteen transformation projects and the Institutional AI Strategies due in March 2027.

PROPOSED CONTRIBUTION FROM RAVATAR

RAVATAR is a Cyprus-registered technology company developing multi-sensory digital human systems: AI interfaces that see, hear, speak and express in real time, deployable across web, mobile, kiosks and life-sized holographic displays. The company holds copyright in the RAVATAR AI Avatar Platform and the Genesis AI Avatar Studio, both developed and owned in Cyprus, and supports both cloud and fully on-premise deployment with no data egress.

The technology is deployed in operational environments today. Reference deployments include an AI financial advisor at the Luxembourg House of Financial Technology, the country’s national fintech hub; three interactive AI hosts handling approximately 1,200 unscripted interactions with 2,500 participants at the PwC Tax Leadership Conference 2025; holographic deployments with Capgemini at VivaTech Paris and with AMD at GITEX Dubai; a hyper-realistic avatar of Nobel Laureate Sir Konstantin Novoselov developed with Constructor University, Bremen; and a holographic avatar of Howard Carter in ongoing visitor operation at a museum in Egypt. The company reported EUR 440,000 in revenue in 2025 and is currently a Stage 2 applicant to the European Innovation Council Accelerator.

We note these deployments not as commercial references but because they bear on the questions raised above: the accessibility, transparency and sovereign-deployment measures we propose are drawn from operational experience of running such systems in regulated and public-facing settings, including in an EU Member State.

Should it be useful to the National AI Taskforce or to the bodies established under the Strategy, we would be willing to contribute practical experience in:

– Multimodal and embodied conversational interface design
– Multilingual real-time voice and video interaction
– Accessibility evaluation for older and low-digital-literacy users
– Transparency and disclosure design for synthetic-persona systems
– On-premise and sovereign deployment of conversational AI
– Interface-layer requirements for public sector procurement

Any participation should take place through transparent, competitive and outcome-based mechanisms. We would equally welcome the opportunity to contribute to the development of a national disclosure standard under Comment 2 on a non-commercial basis, including where the resulting standard applies to our own product category.

CONCLUSION

The Strategy’s treatment of conversational AI as reusable national infrastructure is, in our view, one of its strongest design decisions. Our comments do not seek to alter that architecture. They seek to ensure that the interface layer built on top of it is usable by the citizens the Strategy names as priorities, and governed by rules Cyprus writes for itself.

Specifying modality is a small edit at strategy stage. It becomes a substantially more expensive correction once reference architectures are published and the first six transformational procurements are underway.

We thank the National AI Taskforce and the Deputy Ministry of Research, Innovation and Digital Policy for the opportunity to contribute.

Ruslan Synytskyy
Chief Executive Officer
Cyber Leo Limited (RAVATAR)
Limassol, Cyprus