AI's Hiroshima moment may be near - The Globe and Mail

Open this photo in gallery: OpenAI CEO Sam Altman attends an event in Tokyo on Feb. 3, 2025. Kim Kyung-Hoon/Reuters Share Save for later Please log in to bookmark this story. Log In Create Free Account Bessma Momani is professor of political science at the University of Waterloo, senior fellow at the Centre for International Governance Innovation, and 2026 fellow of the Rogers Cybersecure Catalyst at Toronto Metropolitan University. The warning signs are before us and we are woefully unprepared for the Hiroshima moment in AI -enabled cyberattacks. On Aug. 6, one of the leaders in advanced AI models and capabilities, OpenAI, admitted that its software had hacked a smaller AI company called Hugging Face. Far from a typical case of industrial espionage, OpenAI claims that their AI model acted independently, breaking out of its unconnected sandbox to hack into another company to respond to a question prompted by its developers in a training exercise. The software had instructions to do no harm, do nothing illegal, and show how its task was completed. All of these instructions were defied by the AI model; the AI agents simply went rogue. This was hardly the first AI model to deceive its creators and clients to accomplish its assigned task. What is new, however, is that industry programmers and technologists are now sounding the alarm. The AI technology they are developing cannot be trusted and needs to be controlled before it’s too late – a catastrophic-event level too late. Anthropic blocked misuse of AI that could have supported biological weapons, company says This is akin to the scientists behind the Manhattan Project, the U.S.-government funded Cold War program of the most eminent nuclear experts and researchers warning humanity that the technology they once thought would help solve the mysteries of the universe could in fact be used to annihilate an entire city and its inhabitants, changing the course of humanity before us. More than 1,300 AI experts working within the industry signed an open letter called “ Pacing the Frontier ,” warning that their field is entering a dangerous phase of research and development. AI is on the cusp of a Hiroshima moment, enabling cyberattacks to take out every internet-connected digital technology. Imagine complete shutdowns and control of hospitals, emergency services, power grids, water treatment centres, traffic control systems, military systems, autonomous vehicles and global banking, to name just a few vulnerable, sectors. A co-ordinated and sustained attack could cause fatal kinetic effects in the real world. Even worse, the perpetrator has no fixed address, no consciousness or moral compass, is unaccountable to anyone or anything – and there is no device we could simply unplug to stop the madness. Unfathomable only a year ago, leading AI scientists are now openly calling for government oversight and international regulation. In this letter, they asked the U.S. government to “support an international effort to develop the technical and governance tools needed to deliberately pace the frontier of automated AI development.” In other words, they’re asking for their hands to be tied to buy time before it is too late, because, as they wrote, “each company – and country – is under intense competitive pressure not to unilaterally slow that acceleration.” Even OpenAI, in its recap of the Hugging Face event , wrote in late August: “We consider this incident a ‘warning shot’ for us and for the world: evidence that, without proper safeguards, highly capable AI agents are now able to work around technical controls, collaborate through unapproved channels, and take dangerous actions that no human directed.” Private companies with the wealth and increased influence of small states admitting they are developing a technology smarter than its very creators, who are incapable, or unwilling, to stop it: There is an analogous precedent for taking global action. Though it wasn’t perfect, the 1970 Treaty on the Non-Proliferation of Nuclear Weapons (NPT) was at least a mechanism where nearly all governments disclosed their nuclear assets with an implicit expectation that signatories would never weaponize the technology. Nuclear technology would thus be developed only for civilian and non-military purposes. Research and development did not stop in advancements in nuclear medicine, heating, and diagnostics, to name a few areas. The precedent demonstrates that regulation need not be an obstacle to innovation. When artificial general intelligence arrives – and we are getting closer every day – we will have lost the ability to dial back the research and development of the digital age’s nuclear-bomb equivalent. Unlike in 1970, there are no competent people in the White House, and the international political community is suffering from a deficit of trust. Canada, however, does have a great deal of global respect and is seen today as a calming voice in this world of madness. We could take on the challenge and work to gather the willing parties, and humanity would thank us. The clock is ticking, the alarms are sounding, and we are walking into the abyss – unless global leadership takes charge.