EU AI Act: What Appraisers Are Responsible For When Using AI

As modern software reaches appraisal practices and inspection bodies across Europe, regulatory uncertainty has grown with it. When generative algorithms help formulate condition reports, repair calculations or valuations, the question of legal consequences follows inevitably for colleagues in the field. With Regulation (EU) 2024/1689, known as the EU AI Act, the European Union created what the European Commission describes as the world's first comprehensive rulebook for artificial intelligence.

To keep practical implementation manageable for businesses and to remove existing overlaps, the European legislator published Regulation (EU) 2026/1744 in the Official Journal of the European Union on 24 July 2026. This so-called Digital Omnibus on AI entered into force on 27 July 2026 and modifies central deadlines and definitions of the AI Act. For the independent appraisal profession, this development means one thing above all: clarity for everyday practice.

• 12 July 2024: Publication of the EU AI Act (Regulation (EU) 2024/1689) in the Official Journal of the EU.
• 2 August 2025: The requirements for general-purpose AI models (GPAI) become applicable.
• 24 July 2026: Publication of the Digital Omnibus on AI (Regulation (EU) 2026/1744) in the Official Journal.
• 27 July 2026: The deadline adjustments of Regulation (EU) 2026/1744 enter into force.
• 2 December 2027: New deadline for stand-alone high-risk systems under Annex III of the AI Act.
• 2 August 2028: Deadline for high-risk systems as safety components of physical products under Annex I.

The fear of unwittingly becoming the operator of an over-regulated system by using intelligent office software proves unfounded on a close reading of the provisions. Anyone who favours digital appraisal reports based on structured data stands on solid legal ground, provided that elementary professional principles are observed.

The core of the EU AI Act is a risk-based approach. The strictest requirements — ranging from comprehensive risk management systems through complete data quality checks to conformity assessment procedures by external bodies — apply exclusively to systems with unacceptable or high risk. Annex III of Regulation (EU) 2024/1689 defines this group of high-risk applications in certain sensitive areas, from biometrics and critical infrastructure through education and employment to migration and the administration of justice.

The areas listed in Annex III include, among others, biometrics, the operation of critical infrastructure, education and human resources management, access to essential private and public services (such as creditworthiness assessments by banks), law enforcement and the administration of justice. The preparation of technical expert opinions, condition reports or valuations by independent appraisers is not contained in this catalogue. Appraisal software therefore does not fall under the stand-alone high-risk categories.

• Critical infrastructure: control of transport networks, water or energy supply.
• Employment and personnel: AI-based evaluation of applications or performance assessments.
• Access to services: automated creditworthiness checks and risk classification of private individuals.
• Justice and the courts: AI support for courts in establishing facts or interpreting the law.
• The appraisal profession: technical reports and damage calculations are not listed in Annex III.

In addition, the Digital Omnibus on AI (Regulation (EU) 2026/1744) has moved the deadline for systems under Annex III from the original 2 August 2026 to 2 December 2027. For appraisers producing a court-proof appraisal report, this means relief on two counts: the field falls outside the high-risk classification, and no short-term implementation deadline is pending.

Besides the areas named in Annex III, systems under Annex I of the AI Act can qualify as high-risk if they function as a safety component of a regulated product (for example in machinery or vehicle manufacturing). At this point the Digital Omnibus on AI made a decisive clarification: the concept of a safety component was narrowed appreciably.

A digital element only qualifies as a safety component if its specific intended purpose is to prevent or mitigate risks to health and safety. AI functions used exclusively to assist users, increase convenience, optimise performance, automate processes or carry out quality control do not fall under it following the Omnibus clarification. Writing and text assistance systems in an appraisal practice meet precisely these criteria of a pure assistance and efficiency function.

• Pure assistance function: generating wording suggestions, text modules and summaries without any control authority.
• Efficiency and process optimisation: automatic formatting of findings and matching of part numbers to save time.
• No direct system intervention: the software never intervenes in vehicle control systems or safety architectures.
• Reservation where real danger exists: if a failure of the system directly endangers life and limb, the exclusion does not apply.

One essential limitation must be observed here: the exclusion is not absolute or blanket. If an AI system were used in a form where a failure or malfunction would endanger the health or safety of persons, the classification as a high-risk system remains. In everyday appraisal work, where the physical inspection and the engineering plausibility check remain with the human being, this case does not arise for pure documentation support.

Behind modern writing assistants and analysis tools sit general-purpose AI models, referred to in European law as GPAI. According to the European Commission, the GPAI obligations of the AI Act have applied since 2 August 2025. In professional discussions this occasionally causes confusion: many appraisers wonder whether they must demonstrate documentation and transparency obligations when they subscribe to GPAI-supported industry software.

European law distinguishes between the provider and the deployer, that is, the professional user. The obligations for GPAI models, governed by Chapter V of the AI Act, fall on the developers and providers of those models. They include technical documentation, a policy to comply with Union copyright law, and a publicly available summary of the content used for training. The appraiser who uses a finished application in their practice is legally a user and is not subject to these provider obligations.

This division of roles protects the profession from disproportionate administrative burden. Appraisers need neither audit algorithms nor document model architectures of their own. Their responsibility concentrates on the professional accuracy of the end product they hand over to clients, insurers or courts.

A further aspect that frequently causes uncertainty is Article 50 of the EU AI Act on specific transparency obligations for generative systems. The Digital Omnibus on AI left these transparency requirements substantively unchanged. In addition, generative systems placed on the market before 2 August 2026 must meet the requirements for machine-readable marking of AI-generated content by 2 December 2026.

For the preparation of appraisal reports, Article 50(4) is particularly relevant. This provision requires the marking of text generated or altered by AI where it is published with the purpose of informing the public on matters of public interest. An appraisal report on vehicle damage or a valuation, however, is not a public document but a confidential professional document for a specific client.

• No publication purpose: appraisal reports are commissioned professional documents for insurers, lawyers, courts or owners, and are not addressed to the public.
• Exemption where human control exists: Article 50(4) makes clear that no marking is required where the text has undergone human review or editorial control.
• An accountable person exists: where a natural or legal person assumes responsibility for the publication, the marking obligation falls away.
• Both conditions met: expert appraisal reports satisfy the requirements of the exemption in full.

Because every careful appraiser reads, professionally assesses and releases the software-supported draft before dispatch, the statutory exemption applies directly. An obligation to disclose in a privately commissioned report that individual text modules were digitally pre-structured therefore does not follow from Article 50(4).

The use of modern software in the appraisal profession rests on one principle: there are no fully automated appraisal reports. An algorithm can match data, search records and suggest wording for standard passages. It cannot, however, replace the on-site inspection, the analysis of a damage pattern or the professional plausibility check.

In practice, intelligent assistants take working time off routine tasks and documentation without curtailing professional authority. The appraiser remains the decisive factor at every stage of the workflow — what is known as human in the loop.

By signing the document, the appraiser assumes responsibility for its content. It is precisely this human release that determines the regulatory classification: where a qualified person vouches with their expertise for the accuracy of the statements, the software is a tool. Which liability consequences follow in an individual case depends on the applicable national law and professional rules.

The software platform DIAVAG for appraisers, developed by Hamburg-based CITO GmbH, is designed on the principle of strengthening appraisers in their independence. As a pure B2B platform, the offering is aimed at independent experts who want to continue using their existing tools and digitalise their office processes.

The platform places value on a transparent delimitation: it employs no appraisers of its own and produces no appraisal reports itself. All reports come from the independent appraisers who use the software as a tool. Nor does it sell or lease measuring devices or hardware cases — colleagues work with their own, familiar measurement equipment.

• Pure software solution: SaaS platform from CITO GmbH with no obligation to use proprietary additional hardware.
• Free choice of devices: appraisers use their own existing measurement equipment and test tools.
• Human-in-the-loop principle: no report leaves the system without personal review and release by the appraiser.
• Own brand marks: the principle seals visible on the platform are designed marks of DIAVAG documenting its own working practice. They are not certificates or test seals of third parties, and in particular not an accreditation under ISO/IEC 17020.

The combination of technical drafting support and a mandatory release loop keeps the workflow within the frame the AI Act draws for assistance systems. Since the interplay of European regulation and professional rules continues to develop, individual legal advice is recommended in borderline cases.