The Weekly Brief — Issue 13

A Quebec election audit has found something worth paying attention to if you use AI to understand politics: when researchers repeatedly asked ChatGPT voting-recommendation questions, one recently created AI-assisted election website became the source it cited most often.

That does not tell us how voters are using ChatGPT or whether AI is changing anyone’s vote. It does show why the sources underneath an AI answer matter—especially when the question is consequential.

Plus: Ottawa is funding 10,000 AI-related work placements through Mitacs, Toronto AI-biopharma company Biossil has raised US$153 million in new financing, and a survey of Canadian office workers found one-third of respondents report using AI agents.

Here’s the week.

What happened

Researchers monitoring Quebec’s provincial election have been testing what ChatGPT and Google’s AI Mode tell people who ask election-related questions.

In the latest round, collected September 23, the Media Ecosystem Observatory at McGill’s Centre for Media, Technology and Democracy found that ChatGPT cited lequebecvote.ca in 63% of responses to its voting-recommendation questions.

That was up from 39% in the first audit and 42% the previous week. Across the researchers’ entire question set—not just voting recommendations—18% of ChatGPT citations pointed to the site.

Google’s AI Mode behaved differently. CMTD says it cited the site only twice across the audit so far, and not at all in the latest week.

Lequebecvote.ca was registered in August and says it uses AI to help identify sources, summarize documents, compare information and prepare drafts.

Its operator has since been publicly identified and the site revised. CMTD says Radio-Canada’s Décrypteurs found that some incorrect descriptions of party positions remained after those changes.

This was not a handful of screenshots.

CMTD uses a fixed bilingual question bank covering voting advice, the electoral process, rumours and campaign news. The questions are repeated weekly in separate conversations so researchers can track how answers and sources change.

Across its first three collection waves, the audit recorded 4,909 queries. The researchers also stress that their test does not represent every question Quebec voters ask AI systems, and answers can change over time.

Why it matters

An AI answer is only part of what you are relying on.

You are also relying on the sources underneath it.

That matters especially with elections, where a polished answer might draw from government websites, political parties, news organizations or relatively new sites a reader has never encountered before.

The AI interface can make those sources feel equally authoritative even when they are not.

That does not mean a source is unreliable because it is new or uses AI. It means you still need to ask:

Who produced it?

Where did the information come from?

Can an important claim be confirmed elsewhere?

Élections Québec makes a similar point. It warns that AI tools can produce convincing but inaccurate information and sometimes get election rules wrong. For official information about when, where and how to vote, it directs voters back to Élections Québec.

The 63% figure does not mean 63% of Quebec voters saw lequebecvote.ca, or that 63% of everything ChatGPT says about the election comes from the site.

It means that, in CMTD’s September 23 controlled audit, the site appeared in 63% of ChatGPT responses to the researchers’ specific set of voting-recommendation questions.

Your move

If you use ChatGPT, Google AI Mode or another AI tool to understand an election, do not stop at the answer.

Open the sources.

If the answer describes a party’s position, check the party’s own material and compare it with credible reporting.

If the question concerns voting rules, dates, identification or polling locations, go directly to the official election authority.

And if an AI gives you a recommendation, separate:

What facts is it giving me?

from

What judgment is it making from those facts?

AI can make information easier to find.

It does not remove the need to decide which information deserves your trust.

Ottawa funds 10,000 AI-related work placements

The federal government is providing up to $162 million over five years to Mitacs to support 10,000 co-funded AI-related work placements for post-secondary students, recent graduates and postdoctoral fellows.

The placements will connect participants with Canadian businesses through two parts of the Mitacs AI Advantage program.

One stream is aimed at helping organizations test and deploy AI. The other supports research projects that apply AI across fields including agriculture and health care.

The funding begins in 2026–27. Government of Canada — 10,000 AI-related work placements

An important distinction: these are 10,000 work placements, not an announcement of 10,000 permanent new jobs.

Why it matters: Canada’s AI strategy is increasingly moving from broad commitments into implementation.

For businesses—especially smaller ones—the problem is often not whether AI tools exist. It is whether the organization has the people, skills and capacity to identify a useful problem and put the technology to work.

The Mitacs program is designed to connect those two sides: businesses that need AI capability and younger Canadians who need practical experience.

The next measure is execution: how many organizations participate, what work the placements actually involve and whether those experiences translate into durable skills and adoption.

Toronto’s Biossil raises US$153 million in OpenAI-led financing

BetaKit reports that Toronto-based Biossil has raised US$153 million in an all-equity financing round at a reported US$1-billion valuation.

The OpenAI Startup Fund led the round, and BetaKit says Biossil co-founder and CEO Anthony Mouchantaf confirmed the financing details directly by email.

Other investors include Founders Fund, Golden Ventures, Panache Ventures and several institutional investors.

Biossil describes itself as an AI-native biopharmaceutical company.

Its model is different from simply asking AI to invent a new drug from scratch. The company looks for drug candidates that were previously abandoned or unsuccessful, uses its technology to identify candidates it believes may still have potential, and then acquires or licenses them for further development.

The new capital is intended to expand its clinical-development capacity and add more drug candidates to its pipeline.

Why it matters: A large amount of capital is moving into a Canadian company trying to connect AI with one of the hardest parts of biotechnology: turning promising science into treatments that can survive clinical development.

But financing is not clinical validation.

The size of the round tells us investors are willing to make a large bet on Biossil’s approach.

It does not tell us that its AI has already produced better treatments or better patient outcomes.

Those results still have to be demonstrated.

One-third of CDW survey respondents report using AI agents

A new CDW Canada workplace survey suggests AI is entering a more operational phase for some Canadian office workers.

CDW reports that daily use of employer-approved AI tools rose to 31% in its 2026 survey, while one-third of respondents said they use agentic AI for work.

The report describes agentic tools as systems that can carry out multistep tasks rather than only generate a single answer. Among respondents who reported using agents, most said they accessed them through platforms such as Microsoft Copilot or Google Gemini.

The survey also found a measurement gap.

CDW says 38% of organizations that had adopted AI were tracking specific performance measures to evaluate their AI investments, while close to half either had no such measures or employees did not know whether they did.

There is an important denominator here too.

The survey covered 1,011 Canadian office workers at organizations with at least 20 employees. Respondents used computers in office settings for at least 11 hours a week. It was an online, English-language and unweighted survey.

So the finding does not mean one-third of all Canadian workers use AI agents.

Statistics Canada has separately found that 35.9% of Canadian workers aged 15 to 69 reported using generative AI at work during the previous 12 months. The two studies measure different populations and different things, so their percentages should not be treated as directly comparable.

Why it matters: The workplace AI question is beginning to change.

For many organizations, it is no longer only:

Should employees have access to AI?

It is becoming:

What work should these systems actually do—and how will we know whether they are helping?

Agents make that question more important because the technology can potentially perform longer sequences of work with less intervention between individual steps.

That increases the value of good use cases, clear boundaries and measurement.

It also increases the cost of getting those things wrong.

When an AI answer includes links or citations, it can feel as though the system has proven what it is saying.

Not necessarily.

A citation tells you where the system says information came from.

You still need to ask whether that source supports the claim.

Imagine someone gives you an answer and puts three books on the table.

The books may be excellent sources.

But you still do not know whether the answer accurately represents what is inside them until you check.

AI citations work in a similar way.

A useful source check has three steps:

Open it.

Make sure the page actually exists and says what the AI claims it says.

Identify it.

Who published it? A government agency? A political party? A university? A company? A news organization? An anonymous website?

Cross-check it.

If the claim matters, see whether another credible or original source supports it.

A citation is useful.

It is not the same thing as verification.

The next time an AI tool gives you an answer with sources, choose one claim that matters.

Then:

You do not need to do this for every trivial question.

Use the extra scrutiny where being wrong would matter.

Health.

Money.

Work.

Legal information.

Public policy.

And election information.

The goal is not to distrust everything AI tells you.

It is to know when verification is part of using the tool well.

A calm, practical briefing on the AI developments that matter in Canada—delivered every Tuesday.

That’s the week. See you next Tuesday.

— Padge T.

AI Brief Canada is an editorially independent publication published by Groundshift Advisory Inc. and written by Padge T.

Groundshift Advisory helps Canadian organizations adopt AI responsibly. Sponsored, affiliated, or commercially connected content will always be clearly disclosed.

Issue disclosure: Nothing in this issue is sponsored, affiliated, paid, or commercially connected.

AI use: AI tools supported research, source discovery, outlining and drafting for this issue. All material was reviewed, edited and fact-checked by Padge T. before publication. Read our AI Use Disclosure.

Not affiliated with the Government of Canada.

Corrections or feedback: We welcome corrections and reader feedback. Email [email protected] or review our Editorial Policy.