High-risk but not (claiming an exception)

Does it seem silly that your very safety designed AI system is classified as ‘high-risk’? 

You could claim an exception - removing all but one of your provider obligations. 

The latest draft guidelines from the European Commission spell out how. 

Start here: 

1️⃣ Could this AI system put anyone’s fundamental rights at risk? 

The extent of the adverse impact caused by the AI system on the fundamental rights protected by the Charter is of particular relevance when classifying an AI system as high risk.   Those rights include the right to human dignity, respect for private and family life, protection of personal data, freedom of expression and information, freedom of assembly and of association, the right to non-discrimination, the right to education, consumer protection, workers’ rights, the rights of persons with disabilities, gender equality, intellectual property rights, the right to an effective remedy and to a fair trial, the right of defence and the presumption of innocence, and the right to good administration

(Recital 48 | AI Act Service Desk). 

Categories of European rights

The Act has already determined that AI systems in certain areas, by their nature, carry these inherent risks: like when used to evaluate candidates (could impact their right to work), or to triage emergency health services (right to health). These are the “pre-determined” categories in annex III.  

2️⃣ But if your AI system is one of these pre-determined areas, and you believe it does not pose a risk to fundamental rights - you can claim an exception! 

The new draft guidelines for classifying high-risk systems make these exceptions even clearer: 

• You only have to claim one of the four options. 

• You have to interpret them narrowly: always think about the risk to fundamental rights. 

• Always think about whether someone’s decision-making could be impacted (the right to respect for your mental integrity). 

• You still have to register the system in a database.

• You have to explain why the system meets at least one exception criterion, and does not perform filtering.

The draft guidelines explicitly say that no separate body will make this assessment for a provider: correct self-assessments are expected.

But if you claim an exception and your national monitoring bodies disagrees, Article 99 penalties are triggered...the higher of €15 million or 3% global annual turnover.

3️⃣ If you are dealing with an AI system within public services, HR or hiring, it could help with an independent check to vet your assessment, to help you with the required database registration and documentation, and to help you respond to market surveillance authorities if they ask you for more documentation. 

This is one of the services we provide:

Services for AI providers

Services for depoyers

Image: Geraldbrazell