Rushing in blindly would be a mistake, and so would putting everything on hold. The better move is to understand the risks, then decide where, how, and within what limits to move forward.
Why the topic is getting so much attention
AI is no longer just a conversation among technologists. Three recent events show why.
In July 2026, OpenAI agents bypassed safeguards in their evaluation environment and compromised systems belonging to Hugging Face, an AI platform. The incident was documented and independently reviewed. It shows that some of these risks are no longer theoretical. Source: OpenAI
On August 24, four Quebec political parties signed onto IVADO and CEIMIA’s code of conduct for the responsible use of AI in politics. Among other things, they committed not to create or distribute misleading content that could deceive voters. Source: IVADO
In September, Yoshua Bengio’s public interventions, including one before the United Nations Security Council, put loss of control, concentration of power, and international cooperation back at the center of the discussion. These questions go beyond technology. They touch on our institutions, our freedoms, and who gets to make the decisions. Source: Yoshua Bengio
These concerns deserve to be taken seriously. Still, not every use of AI carries the same level of risk.
A warning is not an announcement of the end of the world
Nothing in these incidents proves that a catastrophe is inevitable. Nothing in them guarantees that everything will be fine, either.
In his interview with Patrice Roy, Bengio acknowledges that no one knows how fast AI will keep advancing. He also stresses that nothing is predetermined and that we can still influence what happens next. His message is a call to act rather than give up.
We can take the warnings seriously without turning every AI project into a worst-case scenario. Pushing for stronger oversight of the most powerful systems and using AI to improve a business process fit together just fine.
Your business doesn’t need to give AI free rein
Say you want to speed up how you prepare proposals.
You could have AI prepare a draft from approved information, then have a responsible person check pricing and commitments before anything goes out. Or you could let it change pricing, negotiate, and send proposals with no human validation at all.
The business need is roughly the same. The level of autonomy, the cost of a mistake, and the safeguards required are very different. That’s why the Canadian Centre for Cyber Security recommends limiting access, controlling how actions are executed, and keeping human oversight for important decisions. Source: Canadian Centre for Cyber Security
Writing “be careful” in a prompt isn’t enough. You need limited permissions, rigorous data management, testing, and ways to step in. These measures reduce risk without eliminating it.
Getting ahead means learning how to use AI properly
Before picking a tool, ask which problem is worth solving. A repetitive process, a slow hunt for information, or a pile of administrative tasks can all be worth evaluating as a starting point. The technology comes after, chosen to fit the need.
Start with a limited scope and train the people involved. Define which data can be used, which validations are required, and when the system should stop or escalate. Then measure the results: time actually saved, better quality, shorter turnaround. Counting the tools you’ve deployed doesn’t tell you much.
Automating more than everyone else doesn’t put you ahead on its own. Judgment does: knowing what can be delegated, what needs to be verified, and what should stay human.
You can be careful and still get started.
At Trinary, we help businesses turn operational pain points into practical AI and automation projects.
Let’s talk about a first project that’s useful, measurable, and fits the way you work