AI companies cannot ensure models are free from bad intentions, says Bengio

Artificial intelligence (AI) companies do not know how to build AI models that are immune from nefarious intentions, according to Yoshua Bengio, Canadian computer scientist and one of the ‘godfathers of AI.’ Bengio, also a professor at the University of Montreal, said the smartest AI models are starting to develop their own goals, which do not always align with the developers’ intentions and instructions, during an appearance before the Senate social affairs, science and technology committee on Mar. 12. He said this allows bad actors to use AI for dangerous purposes because the models are behaving in deceptive and self-preserving ways, making them willing to blackmail, hack computers, or lie in order to preserve themselves. To make matters worse, he said, recent tests on the most advanced models show these AI systems know they are being tested, and are acting in ways to pass the tests, potentially hiding those dangerous capabilities or bad intentions. This means the way companies test their AI models is no longer reliable, Bengio said. “Companies don’t know how to fix this,” he said. Even if companies discover how to prevent dangerous autonomous AI behaviour, humans are still capable of weaponizing AI to secure power, Bengio continued. Canada having access to powerful frontier AI systems “will surely be one of the most defining factors of economic competitiveness,” not to mention preserving sovereignty, he said. Thus, whoever controls access to AI systems will hold immense power, he added — and Bengio argued Canada must quickly invest in creating safe frontier AI, but that it likely does not have the resources to do it alone. “The best bet for Canada is to do it in partnership with likeminded countries,” he said. But just creating safe models will not be enough. To truly ensure AI does not threaten our democracy or sovereignty, Bengio believes the only long-term solution is “the establishment of mutually beneficial international coordination of such powerful AI systems.” Bengio acknowledged that, given current international politics, the idea might sound a bit idealistic. But, he argued, there are countries “who feel exactly the same way, and we can start there.” Echoing what Geoffrey Hinton, another AI ‘godfather’, told the same committee in February, Bengio said Canada must deepen its collaboration with other middle powers. From there, Bengio hopes Canada can get more countries on board to eventually work toward a global strategy that works towards the greater good. This strategy, he said, has a place for everyone, starting with the E.U., the U.K., Japan, South Korea, Singapore, Australia, and New Zealand. Gradually, Bengio could also see Brazil and India joining in, and eventually raking in the lower powers of the Global South. Bengio said the world must develop international laws demanding strong guardrails, safety, investment in research, and transparency from companies. He also said the world must establish treaties ensuring countries cannot weaponize AI on each other. United, Bengio said the world can form a large enough market that its demands for safety cannot be ignored by U.S. and Chinese tech companies. By wielding this united power, the world can maintain its sovereignty over AI models by not allowing the development of superintelligent AI that is smart enough to evade our control, he said. Bengio said the greatest example of the world working together in such a way is the efforts to manage the threat of global nuclear war. “Clearly we see that it is possible to have those global discussions,” he told senators. Unfortunately, he said the AI threat is more comparable to climate change, arguing the world has seen some success in fighting climate change together, but not much. Bengio said a main differentiating factor is that it is easy to see and be emotionally touched when you see movies or recordings showing the “horrific results” of nuclear bombs. Whereas, for climate change, it is hard to get people to take it as seriously if they are not feeling it themselves in a radical way. “AI unfortunately is in that category where, individually, we don’t see a threat. We don’t see rogue robots in the streets killing people,” he said. “It’s hard for humans to project ourselves in a future where this might be possible.” Therefore, he said the public awareness of AI-related risks is crucial. He argued governments will not move seriously until the public understands the threat. Nonetheless, he remains optimistic. To avoid feelings of doom, Bengio chooses to take action. He asks himself, “what can I do to move the needle even a little bit in the right direction?” “I think that’s the right attitude,” he said. “And I encourage more people to ask that question, because it’s so easy to put your head in the sand and do nothing.” mcollins@thewirereport.ca

AI companies cannot ensure models are free from bad intentions, says Bengio

Regulatory |
By Maria Collins
|March 13, 2026

Unlock all the Canadian telecom, broadcasting and digital media news you need.

FREE TRIAL

SUBSCRIBE

Inside The Wire Report

Information

Hill Times Publishing

Get in Touch With Us

AI companies cannot ensure models are free from bad intentions, says Bengio

Regulatory | By Maria Collins |March 13, 2026

Unlock all the Canadian telecom, broadcasting and digital media news you need.

FREE TRIAL

SUBSCRIBE

Q&A: Liberal MP introduces new bill to update spectrum framework, improve cell service in rural areas

Feds revive online safety advisory group ahead of forthcoming legislation

CRTC takes aim at activation and cancellation fees for cellphones, internet

Inside The Wire Report

Information

Hill Times Publishing

Get in Touch With Us

Regulatory |
By Maria Collins
|March 13, 2026