Hi {{first_name}} ,

Finance is the department that exists to know what things cost.

77% of finance teams report they are now using AI, but only 35% say they can measure its return. And only 14% are running it with a strategy. 

I’m no accountant, but those numbers which come from Protiviti's 2026 Global Finance Trends Survey, covered here by Jennifer Kervin, don’t add up.

Forecasting is where most of them are using it, at 76%, up from 58% last year. Risk assessment came next at 67%, followed by automating routine work at 56%.

Protiviti says teams are putting AI to work on bigger data sets, more scenario planning and faster answers for leadership. 

All useful, but hard to put a price tag on. There's no row on a statement that reads "we guessed closer this quarter."

Data security and privacy is finance's top worry for the third year in a row, ahead of planning, analytics and AI itself. It stays at the top because AI keeps pulling in data from inside and outside the company, and the more data you feed it, the more you have to lose.

"Organizations cannot scale AI without confidence in the quality, security and governance of their data," says Christopher Wright, who leads Protiviti's CFO Solutions and Business Performance Improvement practice. He would say that, and he's still right.

For more, including where trade uncertainty is pushing finance's priorities and why 83% of CFOs put cash management in their top three:

→ Read: Finance is adopting AI faster than it can measure it

I mentioned the Hugging Face break-in two weeks ago, when an OpenAI test agent spent four and a half days handing itself administrator access. Jennifer Kervin has the update on what OpenAI is doing about it.

The company has stopped training its newest models while it adds safeguards. No date for starting again.

It also says a model called Astra could have "critical cybersecurity capability," which means the ability to carry out attacks on military or industrial systems.

Chris Lehane, OpenAI's chief global affairs officer, told the Guardian that everyone should get ready for "ongoing, persistent" attacks, and that fending them off will take better models. Which is very good news for a company that sells models.

Matt Hartman, who was acting head of cyber at the U.S. Cybersecurity and Infrastructure Security Agency, says companies will need to "treat every agent as a privileged identity." Its credentials and permissions get the same scrutiny as anyone else's.

An agent is a user with keys, probably a good idea to know what it can open.

The U.K.'s National Cyber Security Centre adds two more. Limit how much an agent does on its own, and make sure a person can stop it immediately, without filing a ticket and waiting.

None of that means waiting on OpenAI.

For more, including what OpenAI wants Washington to require before the next models come out:

→ Read: OpenAI warns AI cyberattacks could become 'persistent'

The patios are on borrowed time, which means conference season. Here are two we're keeping an eye on this month.

YYC DataCon is back in Calgary on Sept. 11 at BMO Centre, squeezed from three days into one, with more than 30 speakers. Four streams: the technical build, how other sectors are using it, leadership including ROI and change management, and governance, privacy and trust.

Co-director Mirna Damergi told Kervin that a lot of Calgary's data teams have turned into AI teams over the past couple of years. She calls that a head start, because those people already know the data underneath.

Her example of what a company pays for skipping that groundwork is a model trained on data nobody cleaned or checked for bias. "It has a time cost," says Damergi. "It has a resource cost, and sometimes it has even a result like an outcome cost."

Damergi's bet is that 700 people in a hallway beats a bigger room. Five days later in Montreal, someone is making the opposite bet.

ALL IN runs Sept. 16 and 17 at the Palais des congrès. More than 7,500 people are expected from over 40 countries, with more than 260 speakers, including Yoshua Bengio, Cohere's Aidan Gomez and Mistral's Marjorie Janiewicz. Germany is this year's country of honour, bringing a delegation of more than 50 business leaders and its own pavilion.

The number I'd hang onto is the split in the room. Organizers put it at roughly 60% AI adopters and 40% AI providers.

That makes it two days of holding what's for sale up against what your own team can keep running once everyone goes home. Worth knowing what that is before you get on the plane.

→ Read: YYC DataCon focuses on what happens after the AI plan gets approved

→ Read: ALL IN 2026 brings Canada's AI ambitions into one room

Some light reading for when the offsite runs long

Saskatchewan tells new data centres to bring their own power Jeremy Harrison, the minister responsible for SaskPower, announced the province's new rules from the construction site of a Bell data centre the rules don't apply to. Anyone building a new one brings their own electricity, more than 30 applications are waiting, and Harrison told reporters only Canadian companies will be allowed to build. That makes five provinces with five different sets of rules, since the four we compared this month don't agree either. In the United States it's now a campaign issue, with candidates in at least 21 races running ads about data centres.

AI risk is not just a technology problem: New research shows why leadership matters more than controls IANS and Artico Search asked 113 CISOs about the next two years, and what separated the hopeful from the worried was how much their bosses understood. 80% of the hopeful group said senior leadership had a fair or good grasp of AI risk, compared with 48% of the worried group. The two also split on who owns AI governance, 74% compared to 40%, and on holding their own budget, 81% compared to 50%.

Hacked before their first coffee: Why new hires risk becoming cybercriminals' favourite target New employees don't know what a normal internal email looks like yet, want to make a good impression, and have just announced the job on LinkedIn. Attackers read LinkedIn. The problem is what companies hand them on day one: a password along the lines of "Welcome2026," emailed in plain text and never changed.

Bill Gates warns world unprepared for AI upheaval Gates wrote 6,000 words arguing the move into the AI era will be one of the roughest stretches in human history and nobody has a credible plan for it. His three risks: jobs that don't come back, with entry- and mid-level roles most exposed; attackers gaining ground faster than defenders can patch; and what chatbots that never push back do to young people. He'd support a plan if one existed and doesn't expect one, because "the geopolitical and economic incentives are pushing too hard to go full speed ahead." How’s that for comforting?

A lot of AI adoption didn't start with a boardroom plan. It started with someone just opening a chat window. 

That's fine until it becomes part of how your company functions. One licence doesn't need a strategy, but a forecasting process does. 86% of finance teams don’t have one.

YYC DataCon and All In 2026 are both coming up this month. If you're going, I'd love to hear what stands out to you.

— 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.