Machine learning is not new — recommendation engines, fraud detection and spam filters have quietly run on it for two decades. What felt genuinely new, almost overnight, was a wave of tools that ordinary employees could open in a browser and simply talk to. Three separate Canadian government sources, writing for three different purposes, all independently anchor that shift to the same few months.
Key takeaways
It is unusual for three separate pieces of Canadian federal output — a security awareness bulletin, a voluntary industry code, and a labour-market research article — to converge on the same historical marker without coordinating. Statistics Canada's own study of employment trends states it as a defined study window: “From November 2022—when generative AI applications started gaining traction following the mass availability of ChatGPT—to December 2025, employment generally grew regardless of potential occupational exposure to and complementarity with AI.” (Statistics Canada, Canadian employment trends in the era of generative AI) The Canadian Centre for Cyber Security, writing purely for security-awareness purposes with no reason to echo StatCan's framing, independently states: “Since late 2022, several LLMs (for example, Microsoft's Copilot, OpenAI's ChatGPT and Google's LaMDA) and services using LLMs (for example, Google's Bard and Microsoft's Bing) have gained the world's attention.” (Canadian Centre for Cyber Security, ITSAP.00.041) And ISED's Voluntary Code of Conduct, published roughly ten months later for policy purposes, opens with the identical trigger: “Advanced AI systems capable of generating content — such as ChatGPT, DALL·E 2, and Midjourney — have captured the world's attention.” (ISED, Voluntary Code of Conduct) Three agencies, three different jobs to do, one shared reference point.
The first article in this series drew the operative line the Cyber Centre itself uses: traditional AI recognizes patterns and classifies existing content; generative AI creates content that did not exist before. Recommendation engines, fraud scoring and spam filtering — the traditional half — had already been running inside Canadian businesses for years by 2022, mostly invisibly, embedded in software nobody thought of as “AI.” What arrived in late 2022 was the generative half reaching ordinary users directly, through a plain chat interface, rather than staying embedded behind the scenes inside someone else's product. That is a shift in who could reach the technology and how, at least as much as it is a shift in the technology's raw capability — and it is why the change felt sudden even though the underlying research had been building for years beforehand.
The clearest evidence that this was a genuine inflection, not routine tech-industry noise, is the speed of the institutional response. Canada's Voluntary Code of Conduct, a real piece of federal policy machinery, was issued within roughly ten months of the moment the Cyber Centre and StatCan both date. The NIST AI Risk Management Framework, released January 26, 2023, arrived within about two months. (NIST, AI Risk Management Framework) Governments and standards bodies do not typically produce new frameworks this quickly in response to incremental progress on an existing, familiar technology category. They produce them quickly when a genuinely new category of risk and capability has just become widely accessible, and both Canadian and American institutions treated late 2022 into early 2023 as exactly that kind of moment.
Knowing where this shift actually came from changes how to think about what comes next. If the change had been a single, unrepeatable breakthrough, the sensible posture might be to wait and see whether it was a one-off. But the pattern the three government sources above describe — an existing research trajectory reaching a wider audience through a new access point — is exactly the kind of shift that tends to keep compounding rather than plateauing immediately, because it is driven by access and adoption dynamics as much as by any single technical ceiling. That is also why every source in this series treats “how fast is this changing” and “is the current state of the technology stable enough to build a decision around” as two different, both legitimate, questions — and why a genuinely current fact sheet, refreshed against primary sources rather than carried forward from memory, matters more for this topic than for almost any other in this series.
Picture two companies in 2021, both quietly using machine learning: one runs a fraud-detection model scoring transactions in the background, the other runs a recommendation engine suggesting products on its website. Neither company's staff interact with either system directly — both operate invisibly, wrapped inside other software, doing the classification half of the Cyber Centre's split. Now picture an employee at either company in early 2023, opening a browser tab and typing a question directly into a chat window, receiving a fluent paragraph in response, with no data science team, no model-training project and no procurement process standing between them and the technology. The underlying research programme behind both scenarios has decades of continuity. What changed between them was not that the research suddenly began; it is that the generative, chat-accessible form of it became something an ordinary employee could reach directly — which is precisely the “everyday work” access point the next article in this series builds on with Statistics Canada's own adoption figures.
Related: what artificial intelligence actually means, AI and Canada’s productivity problem, and, on deciding what to do about a fast-moving technology, the AI strategy & roadmapping hub.
Both mattered, but the sources this article cites converge specifically on the access and attention shift, not on a specific technical claim about model architecture. StatCan's own framing is “gaining traction following the mass availability of ChatGPT” — an availability event, not a description of a single technical breakthrough.
No, and this article does not claim that. Machine learning-based classification systems were already common in Canadian business software well before 2022. What was new was the generative category becoming directly, publicly accessible at scale — the Cyber Centre and ISED both date that specific event, not the invention of AI generally.
Statistics Canada's own adoption series shows continued growth through 2026 — usage roughly tripling between Q2 2024 and Q2 2026 — which the next article in this series covers with the actual figures. This article is about the origin of the shift, not a claim about its current trajectory.
This is one page in a plain-English series on how AI actually works and where it fits in a Canadian business.
