Transparency Requirements Apply From August 2
The European Union has started applying new transparency requirements under Article 50 of the Artificial Intelligence Act.
The regulation itself entered into force in 2024 and has been implemented in stages. From August 2, 2026, companies and public organisations providing or deploying certain AI systems must inform people about AI interactions and disclose the artificial origin of specific types of content.
The rules are intended to reduce deception, fraud, impersonation and the spread of materials that may wrongly be perceived as authentic.
However, claims that every piece of content prepared with AI must now carry a visible label are too broad. The applicable obligation depends on the type of system, the nature of the material, how it is published and whether it has undergone substantive human control.
Users Must Know When They Are Interacting With AI
Providers of systems designed to interact directly with people must ensure that users are informed that they are communicating with artificial intelligence.
The requirement covers chatbots, voice assistants, virtual consultants and other automated interfaces. Disclosure is unnecessary only where the use of AI is obvious to a reasonably well-informed and observant person in the relevant context.
For transport and logistics companies, the rule may apply to automated shipment-tracking services, booking chatbots, customer-support voice systems and virtual freight-forwarding assistants.
The information must be clear and distinguishable and provided no later than the first interaction. Companies should not hide it in lengthy terms and conditions or disclose it only after several stages of communication.
Synthetic Content Must Carry Machine-Readable Marking
Separate obligations apply to developers and providers of systems that generate images, video, audio and text.
Such systems must produce content containing machine-readable marking that enables it to be detected as artificially generated or manipulated.
The technical method may involve metadata, digital identifiers, watermarks or other mechanisms. Solutions must be effective, interoperable, sufficiently robust and reliable in light of available technology and implementation costs.
This obligation primarily falls on AI system providers. A company using a third-party generator to prepare an advertisement or corporate video may not be responsible for the technical architecture of the tool, but it may still have a separate obligation to disclose the artificial nature of the material it publishes.
Not Every Use of AI Requires a Label
The rules contain an exception where an AI system performs an assistive function during standard editing and does not substantially alter the input or its meaning.
Spell-checking, punctuation correction, technical formatting or minor image adjustments will not normally turn a publication into content that must be identified as AI-generated.
The exception cannot be applied automatically to every form of editing. Transparency requirements may become relevant where AI creates new claims, changes the meaning of text, adds objects to an image, imitates a person’s voice or generates realistic footage of an event.
Companies must therefore consider not only whether AI was used, but also how significantly it affected the final material.
Deepfakes Require a Visible Warning
The AI Act gives particular attention to deepfakes.
The term covers AI-generated or manipulated images, audio and video that closely resemble existing or plausibly existing people, objects, places, organisations or events and could falsely appear to be authentic.
An organisation publishing such material must clearly disclose that it has been artificially generated or manipulated.
A hidden technical marker alone is insufficient. Users must be able to see or hear an understandable warning without specialist software or the need to inspect file metadata.
The disclosure must be provided no later than the person’s first exposure to the content.
Transport Companies Are Also Covered
The logistics industry is increasingly using generative AI for advertising, staff training, presentations, customer support and illustrations.
A company may, for example, publish a realistic video depicting a truck crash, warehouse fire, container-handling incident or border inspection. Where the event did not occur and the video could be perceived as documentary footage, it should be disclosed as artificially created.
The same principle may apply to a digital replica of an executive, a synthetic dispatcher’s voice or a realistic imitation of a statement by a public authority, carrier or port operator.
Disclosure is especially important where content concerns accidents, road closures, sanctions, strikes, cargo damage or terminal shutdowns. False information in these areas can quickly influence route planning, insurance decisions and freight prices.
Public-Interest Text Is Regulated Separately
Companies and organisations must disclose the use of AI where it generates or substantially manipulates text published to inform the public about matters of public interest.
Relevant subjects can include safety, public services, legislation, environmental protection, public health and major economic or financial developments.
In transport, examples may include notices about border closures, customs changes, dangerous-goods incidents, railway shutdowns, port disruption or restrictions affecting trucks.
Where such text is automatically generated and published without substantive human review, its artificial origin must be disclosed.
Editorial Control Can Remove the Labelling Requirement
The AI Act provides an important exception for text publications.
AI disclosure is not required where the material has undergone substantive human review or editorial control and a natural or legal person assumes editorial responsibility for its publication.
The review must address the substance of the content. The editor should assess facts, sources, conclusions and reliability and must have authority to amend or reject the text.
Automated spell-checking, superficial reading or a purely formal approval step does not amount to sufficient editorial control.
For news organisations and corporate communications teams, this means that AI can remain a working tool. However, the final material must be genuinely reviewed by a responsible editor where the organisation does not intend to label it as AI-generated.
Advertising and Creative Content Have Limited Exceptions
Deepfakes used in evidently artistic, satirical, creative or fictional works must still be disclosed, but the warning should not interfere with the normal enjoyment of the work.
It may, for example, appear at the beginning or end of a video, in the publication description or in the credits.
The fact that content is advertising does not by itself create an exemption. Where a realistic artificial character appears to be a genuine customer, driver, expert or company executive, the risk of misleading the audience must be considered.
Ordinary AI-generated graphics that do not claim documentary authenticity will not necessarily qualify as deepfakes.
Emotion Recognition Systems Also Require Transparency
Organisations lawfully using emotion-recognition or biometric-categorisation systems must inform the people exposed to them.
This may be relevant to airports, terminals, distribution centres and other locations using cameras or behavioural analysis tools.
Separate provisions of the AI Act already restrict or prohibit certain uses of emotion recognition in workplaces. A warning sign therefore cannot make a system lawful where the underlying use is prohibited.
Logistics companies considering such technologies must separately assess whether their purpose and deployment comply with AI and data-protection rules.
Older Content Does Not Require Retroactive Labelling
Content generated before August 2, 2026 does not have to be labelled retroactively.
The European Commission nevertheless encourages voluntary disclosure where older material could mislead an audience.
A limited transition period applies to AI systems placed on the market before August 2, 2026. Providers of those systems have until December 2, 2026 to comply with the machine-readable marking and detection requirement.
The extension applies only to the provider’s technical marking obligation. It does not postpone the rules requiring disclosure of chatbots, deepfakes and certain public-interest publications.
Companies Can Follow the European Code of Practice
The European Commission has developed guidelines and a voluntary Code of Practice on transparency for AI-generated content.
Signing the code is not mandatory. It nevertheless provides standard approaches to marking and labelling and can help providers and deployers demonstrate compliance.
Organisations that do not use the code must show through alternative measures that they achieve an equivalent level of transparency.
For larger businesses, this may require a formal generative-AI policy, designated responsible staff, records of content origin and checks on external agencies producing advertisements or public information.
Fines Can Reach €15 Million or 3% of Turnover
Breaches of Article 50 can result in administrative fines of up to €15 million.
For companies, the maximum can reach 3% of total worldwide annual turnover in the preceding financial year where that figure is higher than €15 million.
A more proportionate rule applies to small and medium-sized enterprises, for which the lower of the two maximum thresholds is used.
When determining a penalty, authorities must consider the gravity and duration of the infringement, the number of people affected, company size, financial benefits, cooperation with regulators and measures taken to reduce harm.
Enforcement will mainly be carried out by national market-surveillance authorities. The European AI Office and European Data Protection Supervisor will have roles in specific cases.
What Logistics Businesses Should Do
Carriers, warehouse operators and freight forwarders need to identify where artificial intelligence is used in their operations.
Particular attention should be paid to customer chatbots, voice assistants, marketing materials, automated news production, video instructions, training content and systems that analyse people.
Companies should decide who reviews AI-created content, who assumes responsibility for publication and when visible or audible disclosure is required.
Contracts with marketing agencies and software vendors should also be reviewed. Outsourcing content creation does not necessarily release the organisation publishing the material from its responsibilities.
The rules do not prohibit businesses from using artificial intelligence. They establish a principle that customers, employees and other users should understand when they are communicating with a machine or viewing material that could otherwise be mistaken for authentic content.
