Virkkunen: loss of control over AI is among the most serious security risks

The most serious security risk posed by artificial intelligence, according to Commission Executive Vice-President Henna Virkkunen, is what would happen if humans lost control over it. Speaking to the Finnish public broadcaster Yle, she said the technologies are advancing so fast that safeguards are indispensable and that systems must operate within clearly defined security limits, with ultimate control remaining with humans; in some cases models have already escaped their testing environments. She described international cooperation as the most important tool against the risks of AI. She also recalled that Union law requires providers to assess risks on an ongoing basis and to maintain specified safety mechanisms – an obligation that does not apply elsewhere in the world. Under Article 55(1) of the AI Act, providers of general-purpose AI models with systemic risk carry out adversarial testing, assess and mitigate systemic risks at Union level and report serious incidents; loss of control is named as a systemic risk in recital 110, which speaks of “unintended issues of control relating to alignment with human intent”, and in the code of practice for general-purpose AI models.