AI adoption across multifamily and PropTech has outpaced the rules meant to govern it, and one of those rules is now in force.
On August 2, 2026, the transparency obligations under Article 50 of the EU AI Act began to apply, setting specific requirements for how certain AI systems interact with people and how AI-generated content is marked and disclosed.
This article covers what Article 50 requires and where those obligations land in real PropTech products: leasing chatbots, generative listings, synthetic imagery, and biometric features.
The rules can also apply to providers established outside the EU when their AI systems are placed on the EU market, put into service in the EU, or their outputs are used in the EU, depending on the specific role and circumstance.
What Article 50 Actually Requires
The obligations split by role. Providers build the systems; deployers put them to use. Each carries different duties.
| Role | System in scope | What’s required | Product implication |
| Provider | Chatbots, virtual assistants, other interactive AI | Design the system so people are told they’re interacting with AI | Build an AI-interaction disclosure into the entry point |
| Provider | Generative AI (text, image, audio, video) | Mark outputs in a machine-readable format, detectable as artificially generated or manipulated (the EU has published optional labelling icons; the underlying disclosure obligation is mandatory) | Attach machine-readable provenance that survives the content pipeline |
| Deployer | Emotion recognition or biometric categorization | Inform the individuals exposed to it | Surface notice at the point of exposure |
| Deployer | Deepfakes (AI-generated or altered media that looks authentic) | Disclose that the content is artificially generated or manipulated | Flag altered media; assess each edit type case by case |
| Deployer | AI-generated text published to inform the public on matters of public interest | Disclose it’s AI-generated, unless it had human review and editorial responsibility | Keep an auditable human-review and editorial trail |
The marking obligation isn’t unconditional, though. The Commission’s guidance carves out several categories where the requirement doesn’t apply, including source code, machine-to-machine outputs that no person is ever exposed to, and certain assistive standard-editing functions.
That distinction matters for software systems specifically, where AI often runs backstage rather than facing the end user. Not every AI output, in every context, automatically needs a machine-readable marking.
Enforcement sits with national market surveillance authorities, the EU AI Office for systems under its supervision, and the European Data Protection Supervisor where EU institutions are involved.
Where This Lands in a PropTech Product
A handful of common features sit squarely in scope.
Leasing and resident chatbots
Any assistant answering prospect or resident questions needs a clear signal that the person is talking to AI. This is inexpensive to design in and awkward to retrofit once the flow is live.
Generative listing content
AI-generated descriptions and imagery may trigger the Article 50 marking requirements depending on how and where the content is generated and published. Product teams should design for machine-readable provenance rather than assume that every AI-assisted listing field requires the same treatment.
Synthetic imagery and virtual staging
Virtual staging is a direct example. The Commission’s guidance uses an authentic photograph of an empty apartment furnished with AI as an example of partially AI-modified content, which maps almost exactly onto a common listing workflow. The trigger is whether an edit makes content falsely appear authentic; standard editing is treated differently. The product needs a way to classify edit types at upload and attach a disclosure to those that cross into altered-but-authentic-looking, rather than treating all image edits alike.
Emotion or biometric features
Some smart-building access, security, or engagement-analytics tools touch emotion recognition or biometric categorization. The obligation triggers when such a system is deployed on identifiable people. The product needs to surface notice at the point of exposure and log that the notice was shown, so the disclosure is demonstrable rather than assumed.
What Article 50 Means at the Product Level
The obligations translate into a small set of engineering capabilities. Four of them tend to determine whether disclosure is cheap or expensive to meet.
- AI interaction layer. For the interactive-AI disclosure duty: a reliable way to detect when a user enters an AI interaction and surface the disclosure at that point.
- Content provenance. For the marking obligation on generative output: metadata and provenance that survive the content pipeline rather than disappearing when assets move between systems.
- Human review workflows. For the public-interest text exemption: preserve human review as an auditable business process rather than an informal step, since it can determine whether a disclosure is owed at all.
- AI inventory and governance. Underpinning all of the above: visibility into where AI is embedded across chat, search, content generation, analytics, and automation, rather than treating each feature as isolated.
What the New Guidance Clarifies
The European Commission published the final Article 50 guidelines on July 20, 2026, clarifying the scope and application of the transparency obligations, including direct interaction with AI, synthetic content, deepfakes, AI-generated text on matters of public interest, and relevant exemptions.
The Code of Practice on Transparency of AI-Generated Content is also final. On July 8, 2026, the Commission concluded that the Code adequately covers Articles 50(2), 50(4), and 50(5), and the AI Board adopted that adequacy assessment. The Code is voluntary, but adherence is an accepted route for demonstrating compliance with the relevant marking and labelling obligations. Companies that do not follow it must demonstrate compliance through other equivalent means.
The EU has also published official icons for labelling AI-generated content. The icons are optional; the underlying Article 50 disclosure obligations are not. The icons are one way to implement the required disclosure, not a separate requirement.
Why This Connects to the Rationalization Push
The disclosure work gets easier when the stack underneath it is legible. Operators consolidating applications and standardizing on a smaller core have fewer places where AI touches an end user, and fewer surfaces to audit for the signals Article 50 asks for. The same data-foundation work that makes analytics trustworthy also makes it clearer where generated content and interactive AI actually live in a product. Teams already investing in that groundwork are, incidentally, in a better position to absorb transparency rules without a scramble.
Practical Takeaways
- Inventory where AI touches an end user: chat, generated copy, generated or edited media, and any biometric or emotion feature.
- Add a clear “you’re talking to AI” signal to interactive assistants first, since those are the simplest to fix.
- Apply the published EU icons and the Code of Practice so marking work follows the accepted route rather than being rebuilt later.
- Keep a human-review and editorial trail for any AI text published to inform the public.
- Confirm EU exposure with counsel before concluding the rule doesn’t apply.
- Check whether a system predates August 2, 2026: certain AI systems already on the market have a transition period for the Article 50(2) marking obligations, with the relevant deadline extending to December 2, 2026.
Transparency requirements tend to be inexpensive to design in and costly to bolt on afterward. Teams weighing where AI disclosure fits an existing roadmap can treat Article 50 as a reason to map their AI surface area before the next feature ships. At ORIL, we approach these requirements as an engineering problem: mapping where AI enters the product, identifying the workflows affected by transparency requirements, and designing the architecture so disclosures, provenance, and human-review steps can be implemented without reworking the product later.
