A major part of the EU AI Act is about to become real for businesses that use customer-facing chatbots, publish AI-generated material, or create synthetic images, audio, and video.
From 2 August 2026, the transparency obligations in Article 50 of the EU AI Act begin to apply. The rules are designed to help people recognise when they are interacting with AI or viewing certain content that has been generated or manipulated by it.
For small businesses, the message is not that every use of AI needs a warning label. It is that customer-facing AI and some forms of synthetic content now need a deliberate transparency process rather than an improvised disclaimer added at the last minute.
What Changes on 2 August 2026?
Article 50 creates different duties for different roles. Some requirements apply to the provider of an AI system, while others apply to a deployer: an organisation using AI under its authority for a professional purpose.
The key transparency requirements include:
- AI interactions: People should be informed when they are interacting directly with an AI system, such as a customer-service chatbot, unless this is already obvious to a reasonably informed person.
- Machine-readable marking: Providers of systems that generate synthetic text, images, audio, or video must ensure outputs are marked in a machine-readable format and can be detected as artificially generated or manipulated.
- Deepfakes: Businesses using AI-generated or manipulated image, audio, or video that constitutes a deepfake must disclose that the content has been artificially generated or altered.
- Public-interest text: AI-generated or manipulated text published to inform the public on matters of public interest may need a clear disclosure.
- Emotion and biometric systems: People exposed to emotion-recognition or biometric-categorisation systems must be informed, subject to the Act’s specific exceptions.
The European Commission’s announcement of the AI transparency Code of Practice summarises these rules and confirms the August application date.
Does Every AI-Assisted Post Need a Label?
No. The Act does not say that every email, advert, product description, or social-media caption touched by an AI tool must carry a visible AI label.
The specific deployer obligation for text is aimed at AI-generated or manipulated material published for the purpose of informing the public on matters of public interest. Article 50 also provides an exception where the content has undergone human review or editorial control and a person or organisation holds editorial responsibility for its publication.
This distinction matters. A staff member using AI to brainstorm a headline is not the same as a business automatically publishing an AI-written article about public health, elections, finance, public safety, or another issue that may affect public understanding.
Even where a legal label is not mandatory, a voluntary disclosure can still be sensible when AI has materially created or altered content in a way that could mislead a customer. Transparency should be proportionate to the risk and context.
What About a Website Chatbot?
If your website uses an AI chatbot, virtual assistant, voice agent, or automated sales adviser, visitors should understand that they are communicating with AI. A simple message at the start of the interaction is usually clearer than burying the information in a privacy policy.
For example:
You are chatting with our AI assistant. It can answer common questions, but a member of our team can take over if you need further help.
The exact responsibility depends on how the service is supplied and branded. A business using a third-party chatbot may be a deployer, while a business that develops a system or offers it under its own name may also take on provider responsibilities. Contracts with AI vendors should make clear who handles user notices, technical marking, records, updates, and compliance evidence.
Who Handles AI Content Marking?
Machine-readable marking is primarily a provider obligation. If your team uses an off-the-shelf image or video generator, the technology provider may add metadata, watermarks, provenance information, or another technical signal to the output.
That does not mean the business publishing the content can ignore it. Your workflow should preserve those signals where possible. Re-saving, compressing, cropping, or passing media through another platform may remove metadata or weaken technical markers.
A useful publishing check is therefore:
- Has the AI tool marked the output?
- Does the marker survive editing and export?
- Is a visible disclosure also required because the content is a deepfake or falls into another Article 50 category?
- Can the business show which tool created the asset and who approved it?
The New Code of Practice
In June 2026, the European Commission published a voluntary Code of Practice on Transparency of AI-Generated Content. On 8 July, the Commission concluded that the code adequately covers the relevant Article 50 obligations and can help providers and deployers demonstrate compliance.
Signing the code is voluntary, and following it is not conclusive proof that every legal requirement has been met. However, the Commission describes it as an EU-wide compliance tool for organisations within scope. Businesses that do not sign remain responsible for their own compliance approach and should be able to explain and document the measures they use.
For many small businesses that simply use established AI products, formally signing may not be the first priority. The more immediate task is to ask suppliers whether they follow the code and to confirm that their product supports the disclosures and markings your use case requires.
Does This Matter to UK Businesses?
It can. The EU AI Act can apply outside the EU where the output produced by an AI system is used in the Union. A UK business serving EU customers, operating an EU-facing website, or using AI output in services delivered within the EU should assess whether it falls within scope.
UK data protection rules also make transparency important even where Article 50 does not apply. The Information Commissioner’s Office advises organisations using AI-assisted decisions to be open about when and why AI is used and to provide meaningful explanations in the right context. Its guidance on explaining AI-assisted decisions is a useful starting point.
A Practical Readiness Checklist
- Inventory customer-facing AI. List chatbots, voice agents, recommendation tools, automated advisers, and AI features embedded in third-party platforms.
- Map your role. Record whether you are acting as a provider, deployer, distributor, or customer for each system. Obtain legal advice where the role is unclear or the use is high risk.
- Add clear interaction notices. Tell users when they are dealing with AI and provide a route to a human where appropriate.
- Review synthetic content. Identify deepfakes, realistic AI voices, generated video, manipulated imagery, and public-interest text that may require disclosure.
- Check technical markers. Ask vendors how machine-readable marking works and test whether your editing and publishing tools preserve it.
- Update publishing approvals. Require human review, name an accountable editor, and keep a record of the tool, source material, checks, and final approval.
- Update supplier contracts. Clarify who provides notices, maintains technical markers, responds to incidents, and supplies compliance documentation.
- Train the team. Make Article 50 part of your existing AI literacy programme so marketing, customer service, and content staff know when to escalate.
The Bottom Line
The August 2026 transparency rules are not a blanket demand to label everything created with AI. They are a targeted requirement to make certain AI interactions and synthetic content recognisable, particularly where people could be deceived or manipulated.
Small businesses should use the weeks before 2 August to map customer-facing AI, check supplier responsibilities, add clear notices, and document how generated content is reviewed and labelled. The strongest approach is a simple, repeatable process that customers can understand and staff can actually follow.
This article provides general information and is not legal advice. Businesses should obtain advice on their specific systems, markets, and obligations.