AI Act High-Risk Classification: What It Means for IP Tools

The AI Act and High-Risk AI: What Patent Attorneys Need to Check Now

Since August 2, 2026, the core obligations of the EU AI Act (Regulation (EU) 2024/1689) for high-risk AI systems are fully enforceable. For patent law firms deploying AI-powered tools for search, drafting, and portfolio management, one question dominates compliance discussions: do our tools fall into the high-risk category? The answer is more nuanced than most vendors admit.

The Risk Tiers at a Glance

The AI Act divides AI systems into four risk tiers: prohibited systems (Art. 5), high-risk systems (Art. 6-7, read with Annex III), systems with transparency obligations (Art. 50), and all remaining minimal-risk systems.

Two pathways lead to high-risk classification. First: the AI system is a safety component of a product, or is itself a product, covered by the EU harmonization legislation listed in Annex I (e.g., medical devices, machinery). Second: the AI system falls under one of the use cases listed in Annex III - including biometric identification, critical infrastructure, education, employment, law enforcement, and administration of justice.

Patent AI tools do not obviously fit either pathway. But the edges are less clear than they appear.

Why Most Patent AI Tools Are Not High-Risk

An AI system that performs patent searches, drafts claims, or analyzes patent portfolios is neither a safety component nor does it fall under the Annex III use cases. Patent software does not determine access to education, employment, or public services. It does not make decisions about natural persons that affect their fundamental rights.

The conclusion: most patent AI tools fall squarely into the minimal-risk category. No conformity assessment, no CE marking, no mandatory quality management system under Art. 17 required.

This assessment aligns with the European Commission's guidance published in March 2026, which explicitly clarified that professional productivity tools without fundamental rights impact are not to be classified as high-risk.

The Edge Cases Firms Should Know About

Despite the general all-clear, scenarios exist where patent AI tools could drift into high-risk territory. Three constellations matter.

First: if an AI system is deployed to support judicial decision-making in patent disputes. Annex III, Point 8 covers AI systems that assist "judicial authorities in researching and interpreting facts and the law." A tool specifically designed for use by courts or arbitration panels in patent matters could fall under this category.

Second: if a patent AI tool feeds into an employment decision - for example, AI-assisted evaluation of inventor remuneration claims that directly affect the compensation of natural persons. Annex III, Point 4 covers systems used for "employment, workers management and access to self-employment."

Third: if the patent software operates as a component within a larger system that is itself classified as high-risk - for instance, as a module in a regulatory authority's decision-making pipeline.

What High-Risk Classification Actually Requires

Should an IP tool be classified as high-risk, the obligations are substantial. The provider must establish a risk management system under Art. 9, operate a quality management system under Art. 17, create and maintain technical documentation under Art. 11, conduct a conformity assessment (self-assessment under Annex VI is sufficient for most high-risk categories), and affix the CE marking.

Additional requirements include human oversight (Art. 14), accuracy and robustness (Art. 15), cybersecurity measures, and automatic logging (Art. 12). The European Commission's own impact assessment estimates compliance costs for a high-risk system at EUR 200,000 to 400,000 - a significant figure for specialized IP software vendors.

Practical Risk Assessment for Law Firms

Regardless of formal classification, law firms evaluating AI tools should conduct their own risk assessment. The key questions are: Does the tool make decisions with direct legal effect, or does it only support human decision-makers? Does the tool process personal data of natural persons in a way that affects their rights? Is the tool used exclusively internally, or also in proceedings before authorities or courts?

A tool that serves exclusively as a drafting aid or research support for patent attorneys - where the attorney makes the final decision - will almost invariably qualify as a minimal-risk system. That changes the moment the tool acts autonomously or makes decisions with direct external effect.

What Vendors Should Prepare Now

Even though patent AI tools typically escape high-risk classification, vendors must be prepared for classification questions. Clients and compliance departments at large corporations will ask. The AI literacy obligation under Art. 4, enforceable since February 2, 2025, already requires providers and deployers to ensure their staff has sufficient AI competence.

Specifically, vendors should prepare a documented self-assessment of risk classification that traces the analysis through Art. 6 and Annex III in a reproducible manner. They should comply with the general transparency obligations under Art. 50 - particularly the requirement to disclose that content was generated by AI. And they should voluntarily adopt elements of the high-risk framework that constitute good practice regardless: documentation of training data, bias monitoring, and human oversight mechanisms.

Conclusion

The AI Act is not cause for alarm in the patent AI industry. Most IP tools clearly fall into the minimal-risk category. But the edge cases are real, and the ability to present a well-reasoned risk classification will become a hygiene factor in vendor selection over the coming months. Firms that ask these questions demonstrate diligence. Vendors that can answer them convincingly demonstrate maturity.

Want to know how WunderIP addresses AI Act compliance? Talk to us about classification and regulatory readiness.