Mexty and the EU AI Act: Building Trustworthy AI-Powered Learning by Design

On 2 August 2026, important provisions of the European Union’s AI Act become applicable, including Article 50 transparency requirements for certain AI systems and particular categories of AI-generated or AI-modified content. These rules are intended to help people recognise when they are interacting with an AI system or encountering synthetic or manipulated content.

AI literacy requirements are not new. Article 4 of the AI Act has applied since 2 February 2025 and requires providers and deployers of AI systems to take appropriate measures to support the AI literacy of employees and other people using AI systems on their behalf. The measures should reflect their knowledge, experience, training and the context in which the technology is used.

For learning and development teams, this creates an important moment.

AI is rapidly changing how organizations create courses, support learners, assess knowledge and distribute information. But responsible adoption cannot be reduced to adding an “AI-generated” label to a document or publishing a general AI policy.

It requires a complete and controlled workflow.

At Mexty, we have prepared our AI-Native secure Learning Infrastructure around that principle: innovation and governance should not be treated as competing objectives. They must be designed together.

Why the EU AI Act matters for learning platforms

Learning platforms are becoming deeply connected to AI.

An AI authoring tool for L&D can help create course structures, scenarios, questions, explanations and feedback. An AI assistant may guide an author during course development or support learners while they complete an activity. AI can also help transform policies, product documentation and training materials into more accessible learning experiences.

These capabilities offer significant value. They can help teams:

• accelerate course development;
• reuse existing knowledge;
• generate differentiated learning activities;
• provide more contextual support;
• improve accessibility and localization;
• simplify updates;
• create richer opportunities for practice.

However, they also introduce new questions.

Was the content created or significantly modified using AI? Which model was used? What information was provided to the model? Was the output grounded in an approved source? Did a qualified person review it? Can the organization identify what changed between versions? Could the output mislead a learner? Is an AI interaction clearly identified as such?

These are not merely legal questions. They are questions of instructional quality, organizational accountability and learner trust.

The EU AI Act makes some of these concerns more explicit, but responsible learning design should address them even where a specific transparency obligation does not apply.

Transparency requires more than a visible label

Article 50 does not impose identical labelling requirements on every piece of AI-assisted content. Its obligations depend on the AI system, the role of the provider or deployer and the type and context of the content involved.

For example, the rules address informing individuals when they interact directly with certain AI systems, machine-readable marking of AI-generated or manipulated outputs by relevant providers, and disclosure requirements covering deepfakes and certain AI-generated or manipulated text published to inform the public on matters of public interest.

For learning platforms, the broader lesson is clear: organizations need to know where AI is being used and be able to communicate that use appropriately.

Mexty supports this approach through the clear identification of AI-assisted content and interactions.

This can include distinguishing between:

• content written entirely by a human;
• content initially generated by AI and reviewed by a human;
• existing content modified with AI assistance;
• learner interactions with an AI agent;
• AI-generated images, audio, video or other media;
• AI-supported assessments or feedback.

Transparency should help users understand the nature of the experience. It should not become a superficial badge that replaces meaningful governance.

Human review must remain part of the publishing process

AI can create convincing material very quickly. It can also produce errors, omit essential context, misunderstand source documents or generate statements that sound more certain than the underlying evidence justifies.

This is particularly important in workplace learning.

Incorrect content may affect safety procedures, regulatory compliance, customer interactions, employment practices, product use or important business decisions.

That is why Mexty is designed as a Trusted AI authoring platform, not an autonomous publishing engine.

AI-generated material should be treated as a draft until it has been reviewed and approved by an accountable person. Authors and reviewers should remain able to:

• inspect the generated output;
• compare it with the approved sources;
• correct inaccuracies;
• rewrite unclear sections;
• remove inappropriate content;
• change activities and assessments;
• approve the final experience before publication.

Full manual editing is essential. Human oversight is meaningful only when people can genuinely control the result rather than merely accept or reject an entire generated course.

A responsible Secure AI authoring platform must therefore combine automation with practical editorial control.

Grounding AI in approved Sources of Truth

Generic AI models are trained on broad collections of information. That can be useful for general knowledge, brainstorming and language support, but enterprise learning often requires something more controlled.

Organizations need training content to reflect their own:

• policies;
• procedures;
• technical documentation;
• approved terminology;
• products and services;
• compliance rules;
• operational practices;
• brand and communication standards.

Mexty uses approved Sources of Truth to help ground AI-generated learning content in the materials selected by the organization.

This creates an important distinction between asking a general-purpose model to create a course from its general knowledge and using an AI-native platform for creating interactive learning experiences based on controlled organizational sources.

Grounding does not eliminate the need for human review. Source documents may themselves be outdated, incomplete or contradictory. However, it gives authors and reviewers a clearer foundation for validating outputs and maintaining content integrity.

It also makes updates more manageable. When a policy or procedure changes, organizations need to understand which learning content depends on it and whether that content should be reviewed.

Traceability of AI actions, models and changes

Responsible AI adoption requires more than knowing that AI was used at some point.

Organizations may need to understand:

• which AI  function was used;
• which model or provider supported the action;
• what source materials were involved;
• when the action occurred;
• what output was produced;
• what edits were subsequently made;
• who reviewed and approved the content.

This is why traceability, version history and audit trails are central to Mexty’s approach.

Traceability supports several objectives simultaneously.

It helps authors understand how content evolved. It gives reviewers better context during validation. It allows administrators to investigate an issue. It supports continuous improvement and provides evidence that the organization has implemented meaningful governance rather than relying on informal practices.

An Enterprise-ready AI authoring tool should not treat AI generation as an invisible event. It should make AI-supported creation part of an accountable content lifecycle.

AI literacy must become an operational capability

The AI Act’s literacy requirement is particularly relevant to learning and development teams.

AI literacy is not satisfied by sending employees a short definition of artificial intelligence. People need guidance appropriate to their roles and the systems they use.

An author creating compliance training needs different knowledge from a learner using an AI tutor. An administrator configuring a workspace has different responsibilities from a reviewer approving AI-generated media.

Effective AI literacy may include understanding:

• what an AI system can and cannot do;
• how hallucinations and other errors can occur;
• why outputs need verification;
• how approved sources should be used;
• when personal or confidential data should not be entered;
• how to identify inappropriate or misleading content;
• how to report a problem;
• when human judgment must override an AI suggestion.

The European Commission has published practical AI literacy resources and examples to help providers and deployers develop proportionate initiatives.

Mexty’s approach includes guidance for authors, administrators and internal teams so that responsible AI use becomes part of daily working practices not simply a policy stored in a shared folder.

Protecting learners in assessments and analytics

AI introduces particular risks when it influences assessments, learner analytics or decisions affecting individuals.

AI can help authors draft questions, propose feedback and generate scenario variations. It can also help identify patterns in learning activity. But these outputs should not automatically determine whether a learner is competent, eligible for an opportunity or subject to a significant decision.

Assessment questions generated by AI should be reviewed for:

• factual accuracy;
• relevance to the learning objective;
• ambiguity;
• unintended bias;
• appropriate difficulty;
• the quality of feedback;
• whether the assessment measures meaningful capability.

Learner analytics also need context. Completion rates, scores and activity data may inform decisions, but they rarely provide a complete picture of someone’s knowledge or performance.

Mexty therefore places safeguards around assessments, analytics and significant decisions. Human accountability must remain central, particularly where an outcome could materially affect a learner or employee.

The purpose of an AI Agent for Learning should be to support learning and reflection not to make unsupported judgments about individuals.

Reviewing AI-generated media

The transparency debate often focuses on written content, but AI-generated or AI-modified media can also include:

• images;
• illustrations;
• diagrams;
• voice-overs;
• audio;
• video;
• avatars;
• animations;
• captions;
• transcripts;
• thumbnails.

These assets should be reviewed for accuracy, appropriateness, accessibility, licensing, potential bias and the risk that they could mislead learners.

Realistic media deserves particular attention. A synthetic voice, realistic avatar or generated workplace scene may be perceived as authentic even when it represents a fictional person or situation.

The EU has published optional icons and guidance connected to the disclosure of certain AI-generated content. The Commission notes that using an icon does not, by itself, establish legal compliance; the relevant disclosure still needs to meet Article 50 requirements.

For Mexty, media review is part of the broader publishing workflow rather than an isolated compliance step.

Reporting and correcting learner-facing issues

Even with strong controls, issues can occur.

A learner may encounter an inaccurate explanation, inappropriate image, misleading AI response, broken interaction or assessment question that does not reflect the approved source.

A trustworthy Secure interactive learning platform should provide a process for:

1. reporting the issue;
2. recording the relevant context;
3. investigating the content or AI interaction;
4. identifying the source of the problem;
5. correcting or withdrawing the affected material;
6. reviewing whether related content is also affected;
7. documenting the resolution.

This creates a practical feedback loop between learners, authors, reviewers and administrators.

It also turns governance into an ongoing process. Responsible AI is not something completed when a course is published. It continues throughout the life of the learning experience.

Combining governance with interactive learning

Compliance should not force organizations back into static, document-heavy training.

A governed learning environment can still be creative, adaptive and engaging.

Mexty enables teams to create interactive courses without coding, build scenarios and assessments, and deliver learning through a SCORM-compatible environment while maintaining human review, traceability and source grounding.

As an AI-native LMS and authoring platform, Mexty connects creation, review, delivery, learner support, analytics and continuous improvement.

This matters because AI governance should follow the content across its lifecycle. Controls should not disappear once material has been exported or assigned to learners.

The objective is not to slow learning teams down. It is to simplify the instructional design workflow while making the resulting process more transparent and controllable.

From compliance to confidence

The immediate deadline of 2 August 2026 will understandably lead many organizations to review their AI systems, transparency practices and documentation.

But the opportunity is bigger than regulatory compliance.

A controlled AI workflow can help organizations:

• improve the reliability of learning content;
• protect personal and organizational information;
• strengthen accountability;
• respond more effectively to learner concerns;
• update training more consistently;
• increase confidence among authors, administrators and learners;
• scale AI adoption without losing human oversight.

At Mexty, our objective is not simply to help organizations claim that they comply with the EU AI Act.

It is to help them build trustworthy, transparent and human-controlled AI-powered learning.

Innovation and governance should not compete. When both are designed into the same infrastructure, organizations can use AI more confidently while protecting learners, personal data, content integrity and human accountability.

That is the next phase of AI in learning and Mexty is ready to support it.

This article provides general information and does not constitute legal advice. Organizations should assess the AI Act requirements applicable to their specific systems, roles and use cases.

If you enjoyed this, you’ll love our next articles Mexty V3 Is Live: An AI-Native Secure Learning Infrastructure for the Future of Enterprise Learning