2 min

Companies that violate the new regulations could face fines up to €20 million or 4% of their turnover.

The European Commission, the executive branch of the EU, has leaked to POLITICO a draft of its new AI rules. The Commission plans to make the rules public on April 21.

Under the draft proposals seen by POLITICO, the EU could fine companies which use automated tools to surveil people and their behaviour up to four per cent of their annual turnover. This maximum fine is the same as that applies for data privacy violations under GDPR.

The European Commission would focus on uses of artificial intelligence systems for manipulating human behaviour and social scoring. And they would limit others from entering the bloc if they don’t meet its standards.

Critics claim AI reinforces bias

The EC’s proposals come amid criticism of AI software used for “scoring” individuals for the purposes of recruitment, approving financial loans and other important decisions. Critics claim that the AI and algorithms used in such processes can reinforce existing biases.

The EU could also apply fines to companies that develop and sell banned AI apps; providing false information to authorities, or failing to comply with investigations.

According to documents seen by POLITICO, the rules would also include safeguards and prior authorisation for the use of AI in applications considered “high risk”, such as those with implications for health, safety and fundamental rights and freedoms. In some cases, the Commission could even ban some types of AI altogether, according to POLITICO.

Exceptions allowed for the State

The rules carve out an exception allowing authorities to use the tech if they’re fighting serious crime, according to POLITICO. The rules could allow the use of facial recognition technology in public places, for example, if it limits itself in terms of time and geography.

Such an exception most likely aims to appease some EU members who see value in AI. Countries like France, who have struggled with terrorist attacks, want to integrate AI into their security apparatus. However, privacy hawks and digital rights activists oppose such special allowances for the State. These groups have lobbied the Commission hard to ban these uses outright.

In order for the rules to become EU law, the European Parliament must first review and pass them. There is sure to be a fight looming once the parliamentary debate starts.