ANACOM publishes draft organisational guidelines for the implementation of AI literacy

Articles23 June 2026

Artificial Intelligence literacy is no longer merely a best practice for internal awareness-raising. In the context of the Artificial Intelligence Regulation (hereinafterthe ‘AI Act’), providers and those responsible for the use of AI systems must adopt measures to ensure a sufficient level of AI literacy amongst the people who use or operate these systems on their behalf.

In this context, ANACOM has published draft organisational guidelines which set out, from a practical perspective, how organisations should structure their internal AI training and awareness-raising plans.


Training plan: structure and principles

According to the Draft, organisations must define and implement an ongoing AI training and awareness plan, based on three levels: basic, intermediate and advanced. This plan must be integrated into onboarding processes and existing internal programmes, particularly in the areas of cybersecurity, data protection, compliance and ethics.

Training should be modular and tailored to functional roles, responsibilities and levels of risk exposure, and should draw on practical examples relating to the AI systems actually used by the organisation.

The programme should also be periodically updated and integrated into internal governance and regular organisational review cycles.


Foundational level

At the foundational level, theoretical training should cover the basic concepts of AI, its limitations and internal examples of its use, typical risks – including bias and material error – the duty of human validation, internal rules on confidentiality, the protection of personal data and the protection of confidential information or information subject to a duty of confidentiality or legal privilege, as well as the safe use of systems and knowledge of the internal reporting channel.

In the case of AI systems with agent-based functionalities, the training must include practical guidance on the secure delegation of tasks, the system’s limits of action, the identification of situations in which human validation is mandatory prior to the execution of external or persistent actions, and the protection of credentials, access rights and sensitive information in the context of integrations and tools. It must also ensure awareness that automated execution does not remove human responsibility for supervision and the final decision.

The practical component should focus on systems that generate synthetic content, addressing the limitations of these tools, including the possibility of producing factually incorrect content, the need for human verification and validation, and the prohibition on using unverified outputs in external communications or in critical processes.


Intermediate level

At the intermediate level, theoretical training should cover the relevant legal, operational and ethical risks, the criteria for human intervention and enhanced validation, the potential impact on individuals, interests and fundamental rights, as well as alignment with applicable internal policies, including personal data protection, information security and rules relating to confidential information or information subject to a duty of confidentiality.

This level should also include operational concepts of traceability and usage logging, such as recording the use of AI in sensitive processes, citing sources where available, and justifying human decisions when AI output influences decisions with material impact.

In the case of systems with agent-based functionalities, training must enable users to critically assess the system’s proposed actions, understand criteria for human approval and enhanced validation prior to execution, interpret relevant records and logs, and recognise risks of the automatic propagation of errors, inappropriate decisions or improper actions affecting downstream systems or processes. It should also enable users to understand, where applicable, the need to resort to internal suspension or escalation mechanisms in the event of anomalies or well-founded doubts regarding the reliability, appropriateness or contextualisation of the result in the specific case.

The practical component should focus on recognising situations in which specific obligations under the AI Act may be at stake, namely obligations regarding transparency, information or human oversight, when and to the extent that the relevant legal conditions are met, without prejudice to the legal assessment being carried out by the competent internal bodies.


Advanced level

The advanced level is aimed, in particular, at roles with technical, governance or control responsibilities. The training must cover topics such as governance, data quality and bias mitigation, monitoring and auditing, detection of performance degradation, including drift, technical documentation, version control and incident management, including security incidents and relevant failures.

It must also include technical and organisational measures to control the use of content-generation tools, such as access management, logging, safeguards against the exfiltration of sensitive information, and mechanisms for detecting and mitigating outputs that are materially incorrect or inappropriate to the operational context.

In the case of systems with agent-based functionalities, the advanced level must cover, in a manner proportionate to the organisation’s size and technical context, enhanced technical controls, including the minimum necessary permissions and access, segregation of duties, secure credential management, with rotation and revocation mechanisms, the design, configuration and testing of human approval mechanisms, the implementation and periodic verification of mechanisms for immediate suspension or disconnection, such as kill switches, logging and auditing of outputs and actions executed or proposed by the system, as well as a structured response to incidents and near-misses resulting from improper or unauthorised action.

The practical component involves the ability to operationalise the coordination between technical governance, internal rules and applicable legal obligations, particularly with regard to changes to functionalities, the replacement of suppliers, the integration of components and the assessment of operational and risk impact, with decisions and their justifications recorded in the compliance dossier.


Certification and evidence

The Project does not impose an obligation for external certification nor a mandatory individual assessment of AI knowledge. It does, however, allow for the adoption of proportionate mechanisms to assess and monitor the effectiveness of training measures, as tools for risk management, capacity building and continuous improvement, rather than as mechanisms for individual performance assessment.

Organisations must nevertheless be able to demonstrate due diligence and organisational maturity through the implementation of appropriate, structured and proportionate training measures. To this end, it will be important to maintain adequate and proportionate internal evidence of the training and awareness-raising initiatives implemented.


Final note

AI literacy plays a central role in compliance with the AI Act, but also in the prevention of operational, legal and reputational risks associated with the use of artificial intelligence systems.

ANACOM’s Draft Guidelines thus constitute a relevant practical reference for organisations that are structuring or reviewing their internal AI training programmes; these organisations should, as of now, assess their level of preparedness in light of the emerging requirements of the European regulatory framework.

Un paisaje al atardecer con torres de telecomunicaciones en silueta.

Related professionals

LATEST FROM #ECIJA