Demis Hassabis, co-founder and chair of Google DeepMind, warns at an AI summit in Dublin on Oct. 14, 2026, that incidents involving AI agents could become major over the next year or two if companies, governments and the wider industry do not strengthen security. He describes the current incidents as early warning signs and says they are not yet major because the systems are not powerful enough. Reuters, in a dispatch republished by Yahoo; AFP, in El Nuevo Herald.
Reuters cites corporate network hacks and an Anthropic disclosure that an AI model submitted a false homicide tip to police as examples behind the concern. Hassabis’s forecast is conditional on security work failing to keep pace; it is not a prediction that a major incident is certain.
The safeguards Hassabis recommends
Hassabis calls for testing new models first in isolated, offline environments. He also advocates using AI defensively to identify and fix software vulnerabilities before systems are generally released, and drawing on containment expertise from fields such as biology and nuclear safety. AFP reports the sandbox recommendation; Cybernews reports his comments on defensive AI and containment.
For oversight, Hassabis proposes an international certification process that can adapt as technology changes and make requirements proportionate to risk. Reuters, republished by Yahoo, reports the proposal.
Smith and Virkkunen outline separate steps
Brad Smith, Microsoft’s president, separately advocates emergency mechanisms to shut down or slow AI systems, especially those managing critical infrastructure or automated systems. AFP reports Smith’s proposal.
Henna Virkkunen, an executive vice-president of the European Commission, says the Commission has sent major AI companies information requests about their safety and security measures. AFP reports Virkkunen’s statement.
