EU AI Act for PropTech & Real Estate AI
Tenant screening, rental decisioning, and mortgage-linked property valuation AI all fall under Annex III Category 5 of the EU AI Act. Housing decisions affect fundamental rights — this is one of the sectors regulators are expected to scrutinise most closely in the first enforcement wave.
Covered PropTech AI products
Tenant screening and creditworthiness assessment AI
High-risk — Annex III Cat. 5
Automated property valuation models (AVMs) in mortgage workflows
High-risk — Annex III Cat. 5
Rental application scoring and ranking systems
High-risk — Annex III Cat. 5
AI rental price suggestions (landlord tools)
Likely minimal risk
Property search and recommendation AI
Minimal risk
AI for property maintenance scheduling
Minimal risk
Key obligations
Risk management system
Document and mitigate risks specific to housing AI: discriminatory outcomes based on protected characteristics, use of variables that proxy for nationality or ethnicity (neighbourhood data, name-based analysis), and the risk that automated screening systematically excludes protected groups from housing. Housing is a fundamental right — regulators will scrutinise this sector closely.
Data governance
Tenant screening AI is trained on historical rental data that often encodes discriminatory patterns — landlord preferences for certain nationalities, income sources, or family structures that are illegal under housing discrimination law but embedded in historical decisions. Article 10 requires explicit examination and documentation of training data for these patterns.
Technical documentation
Prepare full Annex IV documentation covering the system's intended purpose (the types of rental decisions it supports), the input variables and their justification, training data sources and bias assessment, performance metrics, and the risk management file. This must be prepared before EU deployment.
Transparency to landlord deployers
Estate agents and landlords using your tenant screening AI are deployers with their own obligations. Your instructions of use must clearly explain what the system does, what it cannot assess, known limitations, and the human oversight required. Landlords are not technical users — the instructions must be written to be understood by non-specialists.
Human oversight
Tenant decisions must remain subject to genuine human review. If a landlord or agent is using your AI and the workflow is structured so that AI rejections are automatically final, your system design fails Article 14. The deploying landlord or agent must be able to meaningfully override the AI's assessment — and must be given the information needed to do so.
Common compliance gaps in PropTech AI
Using address or postcode data as a proxy for protected characteristics
Neighbourhood-level variables (crime rates, average income by postcode, school ratings) can proxy for ethnicity or national origin in cities with segregated housing patterns. Article 9 requires identification and mitigation of these proxy risks — they are not neutralised simply because the model does not use ethnicity directly as an input.
Screening based on social media or non-financial data
Some tenant screening tools incorporate social media analysis, psychographic profiling, or behavioural data. These approaches create serious Article 10 (data quality), Article 9 (risk of discrimination), and GDPR (lawful basis, purpose limitation) risks. The AI Act does not explicitly prohibit these inputs, but Article 10 requires justification for each input variable.
No instructions of use for landlord deployers
Most PropTech platforms provide landlords with access to AI screening scores without providing Article 13-compliant instructions of use explaining what the score means, what it does not capture, and how human oversight should work. This creates liability for both the platform and the landlord.
Assuming a 'recommendation' framing avoids high-risk classification
Framing an AI output as a 'recommendation' rather than a 'decision' does not change the classification. If the recommendation is presented in a way that makes rejection of a candidate the default (low score = rejection in practice), the system is functioning as a decision-making system regardless of the label.
Start your PropTech AI compliance documentation
Nytivo generates Annex IV technical documentation for tenant screening and property AI — including data governance records and risk management files.
Start free trialEU AI Act for PropTech — Frequently Asked Questions
Is our tenant screening AI high-risk under the EU AI Act?
Yes. Annex III Category 5 covers AI systems that evaluate the creditworthiness of natural persons or establish their credit score, as well as AI used in decisions about access to essential services including housing. Tenant screening AI that assesses a prospective tenant's suitability, financial reliability, or rental risk is directly covered. If your system's output materially influences whether a person is accepted as a tenant or the terms of their tenancy, it is high-risk.
Does the EU AI Act apply to automated property valuation models (AVMs)?
Automated property valuation models used to assess property value — for example, for mortgage decisioning purposes — are covered under Annex III Category 5 when they are used as a basis for creditworthiness decisions. A standalone AVM used purely for market analysis or by a seller to set an asking price is not high-risk. But an AVM integrated into a mortgage origination flow that influences loan approval or terms is covered. The determining factor is whether the output materially affects a credit or access-to-housing decision about a natural person.
We use AI for rental price suggestions, not decisions. Are we still covered?
Rental price suggestion tools used by landlords or platforms to set asking rents — where the final price is set by the landlord and not automated — are generally not high-risk under the current Annex III text. However, if the platform uses AI to dynamically price rentals and individual applicants are offered or denied access to a property based on automated pricing thresholds (e.g., the system decides whether to accept an applicant at the requested price), that edge case warrants legal review.
What does Article 10 require for tenant screening training data?
Article 10 requires training data to be relevant, representative, and to the best extent possible free of errors and biases. For tenant screening AI, this specifically means: examining whether historical rental decisions in the training data reflect discriminatory patterns (denying tenancies based on characteristics that correlate with protected characteristics like nationality or ethnicity); documenting the composition of training data; and implementing steps to mitigate identified biases. Using historical tenancy outcomes from a market with documented racial or socioeconomic discrimination as training data without bias examination does not meet Article 10.
When must our PropTech AI comply?
For Annex III Category 5 high-risk systems, the full compliance deadline is 2 August 2026. If your system is already deployed before that date, you must bring it into compliance by then. If you are launching after that date, the system must be compliant before first deployment on the EU market. The required documentation — Articles 9, 10, 11, 12, 13, and 14 — takes significant time to prepare properly. Starting at least 12 months before the deadline is recommended for systems that require retroactive documentation.