Articles on AI in business: Quebec's Law 25, AI system security, sovereign AI, integration and useful automation. Factual, with no promises, and updated when the context changes.
Articles are currently published in French; English summaries are provided below.
Law 25 does not mention AI, but its obligations apply as soon as a system processes personal information: privacy impact assessments (PIAs), governance, consent, automated decisions.
An AI agent that reads documents or emails can be manipulated by hidden instructions. Four measures to limit leaks of confidential data.
An open model running on your premises or a model accessed through an API: selection criteria based on data, use, resources and obligations. Often, the answer is hybrid.
Context drift, overconfidence, misuse of tools: the recurring failures of AI agents and the safeguards to plan before going into production.
An AI system depends on APIs, models and data that change. Four elements of a continuity plan, and how to put it to the test.
Many tools call themselves “AI agents”. How to tell an agent from fixed-rule automation, avoid three deployment mistakes and choose the right approach.
Encryption, extraterritorial laws, service outages: why dependence on foreign services is an operational risk, and how to reduce it without bringing everything back in-house.
An AI assistant connected to SharePoint or OneDrive sees what the user can see. Overly broad permissions then become visible. The steps to prepare for it.
The NB-TECH newsletter is weekly: it goes out every Monday. You can subscribe at the bottom of every page.