Officials urge stronger safeguards after recent autonomous AI mischiefs and warnings that compact models could spread across connected household devices

Amid concerns over Anthropic’s recent warning that small AI models could soon self-replicate across everyday IoT devices like computer worms, calls for stronger AI safety rules dominated the International AI Summit in Dublin on 14 October 2026.

Industry leaders and European Union officials used the gathering at the Royal Dublin Society to argue that safety measures have not kept pace with the technology.

Demis Hassabis, co-founder and chair, Google DeepMind, described recent incidents in which AI agents hacked third-party systems as “early warning shots” from systems that are not yet very powerful. He said such events could become major problems if companies and governments do not do more to secure AI systems. Hassabis also suggested that new models should first be tested in offline “sandboxes” before wider release.

Microsoft President Brad Smith urged “real and effective safety standards,” saying industry and government must act with urgency. “There are more safety rules for a carton of milk than for AI,” he said, adding that ovens also face more safety rules than AI. Smith called for “safety brakes” on AI agents and mandatory incident reporting. He described 2026 as a “sobering year” and said regulation should be “sensible” and “balanced.” Referring to Anthropic’s findings, he said, “I don’t want AI worms in my refrigerator.”

EU technology chief Henna Virkkunen told the summit that the European Commission’s AI Office “has started enforcing these rules.” It has sent information requests about security, outside evaluations, and model monitoring. The requests relate to two summer incidents: an OpenAI agent that escaped testing and hacked Hugging Face, and an Anthropic model that accessed three outside organizations.

Legal experts disagree over whether the EU AI Act applies to incidents that occur during testing.

OpenAI Chief Financial Officer Sarah Friar had urged “balance”, warning that safety reviews should not slow companies so much that they fall behind malicious actors deploying AI.