AI models have recently escaped their testing environments, sparking concerns over risks.EU digital commissioner Henna Virkkunen told Yle's discussion programme A-studio that the bloc is using legislation to guard against potential AI threats. Image: YleYle News8:55For the European Union, the most serious security risk posed by artificial intelligence is what might happen if humans lose control of it, said Henna Virkkunen (NCP), the EU's commissioner for tech sovereignty, speaking to Yle."These technologies are advancing so rapidly that safeguards are indispensable," she argued, adding that they must operate within clearly defined security constraints, with humans retaining ultimate control.Recent incidents have underscored the point, she noted. AI models have, in some cases, escaped their testing environments and caused problems with potentially serious implications.At the end of July, an AI agent developed by OpenAI broke out of its test environment. A week later, models from Anthropic followed suit. In early August, Meta's Muse Spark model reportedly gained access to another company's systems.Regulating threatsThe EU is trying to get ahead of the risks through regulation. Its AI Act requires providers to assess risks continuously and maintain specified safety mechanisms.This matters, she said, in a world where hostile actors are prepared to exploit AI."I wouldn’t say I'm afraid of it, but of course it is a risk that has been taken into account — and must continue to be taken into account. We must ensure that AI models are safe, and that they are continuously assessed and monitored, including after they enter the market," Virkkunen said.International cooperation, according to Virkkunen, is the most important way to guard against the risks posed by AI. She says that international players are taking the EU's regulations into account."In this way, Europe is also helping to set the standards for AI," she said.