EU Says AI Act Can Address Risks From Rogue AI Agents

Summary

European Commission Executive Vice-President Henna Virkkunen said the European Union’s AI Act is capable of addressing risks from highly capable models and potentially rogue AI agents. The risk-based law, adopted two years ago, covers the full lifecycle of models, while regulators are developing guidance for companies on model evaluation and the time needed for safety testing. A scientific panel of 60 experts is expected to contribute recommendations. Virkkunen rejected criticism that the rules are already outdated, saying they require risk assessment, outside expertise and continued monitoring as technology develops. The Commission sent requests in late August to more than 30 AI companies seeking information about safety, security, transparency and copyright practices; the requests could lead to investigations and fines of up to 7 percent of global annual turnover. The companies were not named, but Virkkunen said Chinese AI startups also received requests. She said the most capable models currently on the market come from the United States and China and therefore create new risks that regulators must monitor. Decisions on a possible EU digital levy remain with member-state governments.