EU issues guidelines clarifying banned AI uses

The EU has issued official guidelines on AI applications banned by the AI Act – including real-time biometric identification, social scoring and harmful manipulation.

The 135-page document was published on Tuesday by the European Commission to increase legal clarity for businesses and organizations for AI systems posing “unacceptable risks” to fundamental rights. And while the guidelines are non-binding, they contain legal explanations and practical examples that will help with compliance with the most comprehensive AI regulation in the world.

The ban on AI with “unacceptable risks” listed under Article 5 of the AI Act went into force on February 2nd. The AI Act, however, has left a number of carve-outs for law enforcement agencies and border control to use banned AI tools, including the use of live facial recognition in public spaces in cases of serious crimes such as terrorism.

While using the technology is still prohibited when it comes to identifying a shoplifter, police can seek authorization to use live facial recognition when it comes to a defined set of serious criminal offenses, according to the guidelines. The document also clarifies that police cannot deploy real-time facial recognition technologies in a broad, untargeted manner, for example locating potential criminals by setting up cameras during a busy festival in the city.

Another AI practice banned by the EU’s rulebook is the untargeted scraping of online images and CCTV footage to develop facial recognition databases. The guidelines note that the ban doesn’t apply if the scaping is targeted, for example when searching images for a specific criminal or to identify a group of victims. The prohibition also doesn’t apply where no AI systems are involved in the scraping.

The AI Act prohibits AI systems from inferring the emotions of a person at a workplace and education institutions but leaves exceptions for medical or safety reasons. Identifying if someone is smiling, for example, is not emotion recognition – concluding that a person is happy is.

Similarly, identifying whether a person is sick or tired or any other “physical state” is not emotion recognition which is why AI recognition systems that detect professional driver’s fatigue are not banned.

EU members will have until August 8th to set up market surveillance authorities that will enforce the legislation. The bloc plans to roll out other restrictions, including for “high-risk” AI systems in the following months. Companies can expect hefty fines for certain AI practices: Up to 35 million euros (US$38.1 million) or seven percent of worldwide annual turnover, whichever is higher.

And while many crucial details of the AI Act still remain unresolved, the regulation may see resistance from Big Tech.

Trump’s tariff threat complicates EU’s AI Act implementation

U.S. technology giants are receiving support for their interests in Europe from President Donald Trump. Brussels is worried that its trans-Atlantic partner will pressure the EU around the AI Act to ensure better treatment, according to the Financial Times.

Trump has been criticizing efforts to regulate AI, revoking a 2023 executive order signed by Joe Biden on reducing the risks that the technology poses to consumers, workers and national security.

Big Tech companies such as Meta, Google and Apple have also been opposing EU digital regulations, which aside from the AI regulation include the Digital Services Act. A big thorn in the side for U.S. tech firms is the requirement of more transparency for data, including allowing third parties access to the code of AI models to assess risks.

The EU is now worried that Trump will raise pressure on the EU around the AI Act to ensure smooth sailing for U.S. companies, Patrick Van Eecke, co-chair of law firm Cooley’s global cyber, data and privacy practice, told the paper.

Many Big Tech firms, however, are already in compliance, according to a person with knowledge of the matter.

Article Topics

AI Act  |  biometric identification  |  EU  |  facial recognition  |  law enforcement  |  legislation  |  real-time biometrics