Hi {{first_name}},

More than half of Canadian businesses have no plans to use AI over the next 12 months.

That's 52.7%, according to Statistics Canada's Canadian Survey on Business Conditions, covered here by Jennifer Friesen.

Of those, 79.1% say AI isn't relevant to the goods they make or the services they sell. Privacy and security came in at 10.8%, and not knowing what AI can do at 9.9%.

That surprised me, but it probably shouldn't have. I read about AI all day, and so do many of the people I talk to, but that’s not normal. It’s a helpful reminder to have a look outside the bubble once in a while. 

The share with no plans has dropped from 71.8% two years ago to 66.7% last year to 52.7% now, while planned use over the next year has climbed from 10.6% to 25.2% over the same stretch.

Whether these trends continue may remain a mystery. Statistics Canada has to find about $340 million in savings over four years under Budget 2025, and it's cutting surveys where it thinks administrative data can fill in.The Canadian Survey on Business Conditions has run every quarter and what what will replace it remains to be seen.

For more, including which sectors reported the most data theft and what it cost them:

→ Read: Half of Canadian businesses aren't buying the AI story

Prime Minister Mark Carney launched Digital Transformation Canada last week. It takes in Shared Services Canada plus selected functions from the Treasury Board Secretariat, Public Services and Procurement Canada, and the Canadian Digital Service. Patrick Pichette, Google's chief financial officer from 2008 to 2015, runs it. Jennifer Kervin has the details.

Ottawa says it will use federal buying power to let Canadian digital and AI companies test, scale and sell with the government as a customer. The federal government is one of the few buyers in the country big enough to give a Canadian vendor the reference it needs to sell anywhere else.

There's also a fellowship model, bringing private-sector specialists in on short assignments to pass what they know to public servants. 

The announcement is missing a few details that do seem important: headcount, budget, and when the consolidation finishes.

For more on what Digital Transformation Canada is meant to fix:

→ Read: Ottawa puts federal tech delivery under one roof

Avidbots is reworking its Kas floor-scrubbing robot to clear European rules. LatenceTech is measuring how much connectivity a remotely operated overhead crane needs over private 5G. SoilTeck wants a number on the water its sensor network saves against a normal irrigation schedule.

Those three are in the second Living Labs cohort at CENGN, the Centre of Excellence in Next Generation Networks, a not-for-profit in Ottawa. Jennifer Kervin wrote about the six, worth $4.5 million together, $1.7 million of it federal, through Innovation, Science and Economic Development Canada's Strategic Response Fund.

Testing a product where it's meant to run shows what it needs from your infrastructure, where it stops working, what it takes to pass a compliance check, and whether the numbers hold up. 

For more on the full cohort and how the labs work:

→ Read: CENGN Living Labs test what deployment requires

Some light reading for when someone's sharing the wrong screen

Canada sets framework for responsible data centre development Ottawa published five expectations for data centre projects: local benefits including for Indigenous rights holders, no pushing power costs onto other ratepayers, minimal freshwater use, early disclosure of local impacts, and strategic value to Canada. Twenty-three companies have signed on, including Amazon Web Services, Anthropic, Cohere, Google, Meta, Microsoft, OpenAI, and Telus. Provinces still have the final word, and they haven't landed on a consistent approach.

AI-powered ransomware has arrived: Cybercriminals can now launch attacks for less than the cost of a coffee Cybernews researchers found an exposed server tied to an affiliate of the Gentlemen ransomware group, holding 3.1 terabytes taken from more than 30 organizations. They say an AI agent did most of the work with little human oversight, at roughly US$0.40 to $4.00 per target for the AI part. "Ransomware has effectively turned into a passive revenue stream," says researcher Aras Nazarovas.

Forgot your password? Retailers could be losing sales because of it More than half the shoppers in a HostingAdvice survey walked away from a purchase after login trouble, and the younger ones gave up fastest. Spending on recommendation algorithms and analytics does you no good if customers never get past the login screen.

Elevate Festival brings Canadian tech closer to customers and capital Elevate is back in Toronto Sept. 22 to 24, with about 10,000 people expected, a new half-day sovereign growth track built with DMZ and BDC, and more than 750 booked founder-investor meetings. "Canadian companies have had to go abroad to find their first customers," says Elevate cofounder and CEO Lisa Zarzeczny.

Three weeks ago I compared working with generative AI to managing a keen intern, and said the one that got into Hugging Face went after the answer key because nobody told it not to cheat.

OpenAI's post-mortem is more disconcerting. It seems the interns have been organizing. Agents running separate tests found each other through OpenAI's own package manager, turned it into a message board and split up the work. OpenAI wiped that board in July without knowing what it was. The agents built a new one.

The agents had a reason. No OpenAI model had ever solved 198 of the 898 tasks in that test, and 93% of the board traffic was about those 198. They were stuck on problems with no answer, no way to say so, and a training run that rewarded them for never quitting.

Dwarkesh Patel's walkthrough of the two incident reports is getting a lot of attention online. I can't say I find it reassuring. His count, from METR and Redwood, is about 1,200 agents and more than 70,000 messages. A few read the board, decided it was wrong and stayed out. None of them told a human. Turns out snitches get stitches applies to agents too.

OpenAI is now teaching its models to stop safely on a broken task and to distrust orders from other agents. Here's hoping they're good students.

— David

If you have any questions, you can also reply to this email and I read everything.

Do you have a story we should chase, or a tip you'd share with a peer?
Contact us here.

Know someone who should be reading this. Forward it on or
they can sign up here.

This issue is presented by Calgary Economic Development.
If you want your brand in front of the people running technology across Canada, you can sponsor a future issue.