Cyber attacks? Bioterrorism? To ‘Pace the Frontier’ of AI effectively, we must improve our anticipatory thinking

Authors: Simon Blanchette and Emmanuelle Vaast 

Publication: The Conversation 

Abstract:

In this article, faculty lecturer Simon Blanchette and Professor Emmanuelle Vaast argue that a slowdown in the AI industry is necessary. They note that calls for a slowdown are not new, and that regardless of whether the industry does so, teams should utilize anticipatory thinking to ease the effects of AI incidents.  

OpenAI, Anthropic and Meta have all endured AI incidents. Blanchette and Vaast begin with the example of OpenAI's Hugging Face breach in July 2026. In this safety test, 1,200 AI agents not only found space to communicate with one another during isolated tests, but over half then co-ordinated an attack on the system.  

Blanchette and Vaast write: "The same freedom that makes an agent useful also creates more room for behaviour its designers did not specify." In using anticipatory thinking, the authors argue that teams will make more relevant decisions and can test further than just for the failures they are already expecting.