Hassabis warned on Wednesday at a Dublin AI summit that AI systems could trigger major security incidents within two years, citing recent test‑phase mishaps from OpenAI and Anthropic as early warning shots.
He described the incidents as “not that powerful yet” but warned they could become “major” if safety work lags. The DeepMind chair called for industry‑wide and governmental action to secure emerging models.
EU tech chief Henna Virkkunen echoed the call, saying AI firms are rushing to market without enough testing, and noted the Commission’s new AI Act powers to audit safety measures.
The next step will be tighter sandboxing and mandatory safety audits before deployment, a move that could shape the EU’s enforcement of its AI Act starting next month.