EU AI Act

Global Language Institute (DEMO) EU AI Act Statement

Last updated: 10 November 2026

The EU Artificial Intelligence Act (Regulation (EU) 2024/1689, the "AI Act") sets rules for AI systems placed on the market or used in the European Union. This statement explains how Global Language Institute (DEMO) complies with the AI Act, how learners are informed that they are interacting with AI, and how AI-generated content is labeled. It is intended for learners, schools, businesses, and procurement and compliance teams.

Global Language Institute (DEMO) runs on the Talkio AI language practice platform, developed and operated by Aidia ApS, Denmark.

Summary

Global Language Institute (DEMO) is an AI system for language practice. It is not a prohibited AI practice and is not a high-risk AI system under Annex III.
It is subject to the transparency obligations in Article 50 of the AI Act, which apply from 2 August 2026.
Learners are told that they practice with AI tutors, and organizations can show a one-time AI transparency notice that each learner confirms.
AI-generated text, speech and images are marked in a machine-readable way where our model providers support it, and AI-generated content is kept separate from the learner's own content.
We do not develop, train or place on the market general-purpose AI models. The obligations for general-purpose AI model providers therefore rest with the model providers we use.

Our role under the AI Act

Global Language Institute (DEMO) is an AI system built on general-purpose AI models from established third-party providers. Aidia ApS is the provider of the Talkio AI system. Schools, businesses and other organizations that make Global Language Institute (DEMO) available to their learners act as deployers.

We are not a provider of general-purpose AI models. We do not develop, pre-train or fine-tune foundation models, and we do not make such models available to others. The obligations in Chapter V of the AI Act for general-purpose AI models, such as model technical documentation, a copyright policy, the public summary of training content and the additional obligations for models with systemic risk (Articles 53 to 55), apply to the companies that develop those models, not to Global Language Institute (DEMO). We select providers that publish how they meet these obligations, and we review our choice of models as their documentation and the regulatory guidance evolve.

Customer content, conversations and voice data are not used to train AI models. The current list of sub-processors, including AI providers, is published in the Privacy Policy.

Risk classification

No prohibited practices (Article 5). Global Language Institute (DEMO) does not use manipulative or deceptive techniques, does not exploit vulnerabilities, does not perform social scoring, does not infer emotions, and does not perform biometric identification or biometric categorisation. Pronunciation analysis evaluates how a spoken sentence matches the target language. It is not used to identify, verify or categorise people.

Not a high-risk AI system (Article 6 and Annex III). The intended purpose of Global Language Institute (DEMO) is self-directed language practice with formative feedback. It is not intended to:

determine access or admission to education or vocational training;
evaluate learning outcomes as the basis for grades, certificates or other decisions about a learner;
assess the level of education a person should receive or be able to access;
monitor or detect prohibited behaviour during tests;
recruit, select, evaluate or monitor workers.

Scores, levels and assessments shown in Global Language Institute (DEMO) are indicators to help learners practice. Under our Terms & Conditions, organizations must not use Global Language Institute (DEMO) as the sole basis for assessment or grading without meaningful human involvement.

Global Language Institute (DEMO) is therefore an AI system with transparency obligations under Article 50.

Informing learners that they interact with AI (Article 50(1))

Global Language Institute (DEMO) presents itself as an AI language tutor, and learners are made aware that conversation partners, voices, feedback and assessments are generated by AI.

AI transparency notice. The web app can show a one-time notice on the home screen stating that the tutors are AI systems, not real people, and that their replies, voices, feedback and corrections are AI-generated. The learner confirms the notice and the time of confirmation is recorded. The notice cannot be declined. It is enabled by default on Talkio, and organization administrators can turn it on or off in the Learning Hub organization settings, on the App Settings tab.
Mobile app. Before first use, the Talkio mobile app explains that conversations are powered by third-party AI providers and asks the learner to agree before continuing.
Tutor personas. Tutor names, personalities and the default tutor portraits are fictional and AI-generated. They do not depict real people, even when a portrait looks like a photograph. Organizations that create their own tutors are responsible for the names and images they choose.
Legal terms. Our Terms & Conditions state that tutors are AI systems and that AI-generated content may contain mistakes.

Machine-readable marking of AI-generated content (Article 50(2))

Article 50(2) requires providers of AI systems that generate synthetic text, audio, images or video to mark the output in a machine-readable format so that it can be detected as artificially generated.

Global Language Institute (DEMO) generates the following types of content:

Text (tutor replies, feedback, suggested improvements, translations, summaries, assessments, quizzes and course content). Text generated with supported OpenAI models carries OpenAI's invisible text watermark, which we have enabled for our account. The watermark is embedded in the wording and does not add visible or hidden characters.
Speech (the tutor's voice and read-aloud audio). Speech is played live inside the app and is not provided as downloadable audio files. Speech generated with OpenAI voices carries an inaudible SynthID watermark embedded by the provider.
Images (course covers, content images and tutor portraits created in the Learning Hub). Images generated with OpenAI image models carry an invisible SynthID watermark embedded by the provider. Generated image files are also named as AI-generated.
Video (video tutors, where available). Video tutors are streamed live inside the app and are not provided as downloadable video files.

All AI-generated content is delivered inside an interface that discloses the use of AI, and is stored separately from content created by the learner. We continue to extend machine-readable marking to further output types and providers as the provider tools and the technical standards under the Code of Practice on Transparency of AI-Generated Content mature.

Short text, such as a single chat reply, leaves little room for a statistical watermark. The Code of Practice on Transparency of AI-Generated Content does not require text watermarks in outputs shorter than about 200 tokens.

AI-generated versus AI-modified content

Global Language Institute (DEMO) keeps a clear distinction between content generated entirely by AI and the learner's own content that AI has corrected:

AI-generated content, such as tutor replies, is stored and shown as messages from the tutor.
Corrections and suggested improvements of a learner's typed or spoken sentence are attached to the learner's own message and shown separately from the tutor's replies. In the web app, the changed words are highlighted next to the learner's original sentence, so the learner can see exactly what AI suggested and what they wrote themselves.
Corrections made by the learner, for example to fix a speech transcription, are recorded separately from AI suggestions.

Corrections of a learner's own sentence are an assistive editing function. Article 50(2) does not require machine-readable marking where an AI system performs an assistive function for standard editing or does not substantially alter the input. We still present these corrections as AI suggestions.

Human oversight and limitations

AI-generated replies, feedback and scores can contain mistakes. They are advisory: learners decide whether to follow a suggestion, and no AI output in Global Language Institute (DEMO) automatically triggers a decision about a learner. Organization administrators and teachers can review usage and progress in the Learning Hub, subject to the organization's settings and the learner's consent where required.

We test changes to prompts, models and feedback before release, monitor errors in production, and can switch to other models if a provider becomes unavailable.

What organizations should do

Organizations that deploy Global Language Institute (DEMO) for their learners are deployers under the AI Act. To support their obligations, organizations should:

inform learners that they practice with AI, for example by enabling the AI transparency notice;
not use Global Language Institute (DEMO) scores or assessments as the sole basis for grades, admissions or other decisions about learners;
ensure that staff who manage Global Language Institute (DEMO) have sufficient AI literacy (Article 4);
only use names, photos or voices of real people for custom tutors with those people's consent, and disclose such content as AI-generated.

Contact

For questions about this statement or AI Act compliance, contact us at contact@gli.demo.