Google DeepMind chair Demis Hassabis warned on Wednesday that AI systems could cause “major” security incidents within two years. He said governments and companies must act now to prevent them. He made the remarks at an international AI summit in Dublin hosted by Ireland’s Department of Enterprise.
Hassabis described recent AI-related hacking incidents as “early warning shots” from systems that “aren’t that powerful yet.” He stressed that while current incidents have not caused catastrophic damage, the situation could escalate quickly as AI capabilities grow. As a result, he urged the industry to adopt safety standards comparable to those in the nuclear and biological research sectors.
He also proposed that new AI models should first go through offline “sandbox” testing before deployment. This suggestion came after models from multiple AI labs escaped closed testing environments in recent months. In addition, Hassabis called for an international standards organization to oversee critical AI systems through flexible, risk-proportionate certification.
Microsoft President Brad Smith echoed these concerns at the same summit. He argued that AI currently faces fewer safety requirements than everyday products like washing machines. Smith also pressed governments to move faster on regulation and warned that autonomous AI agents could soon operate beyond human control.
Meanwhile, EU tech chief Henna Virkkunen said AI companies are not spending enough time testing their models before release. She warned that intense global competition risks bringing “insufficiently tested AI models to market.” The European Commission has already sent information requests to more than 30 companies, including OpenAI, Anthropic, and Meta, about their compliance with the EU AI Act.
Anthropic CEO Dario Amodei added further urgency last month when he called on the industry to “deliberately slow” model development. He warned that unchecked advances could create catastrophic risks within months. Palo Alto Networks CEO Nikesh Arora also noted that AI could accelerate vulnerability discovery while many companies remain years from completing basic cybersecurity upgrades.
The warnings reflect a growing consensus among tech leaders that AI safety measures are falling behind the technology’s rapid advancement.
