Aussie firms cautioned on artificial intelligence
Kaaren Morrissey |
Companies deploying artificial intelligence across their systems should take a cautious approach and treat the agents as “insider risks”.
At the same time, firms need to protect against outside threats, such as he autonomous OpenAI agent that gained unauthorised access to Australian government and public systems including health statistics, in June.
“That’s front and centre for all of us,” the head of one of the world’s most influential computing companies told a conference on Wednesday.

Microsoft chair and chief executive Satya Nadella appeared in a pre-recorded video chat with Commonwealth Bank of Australia boss Matt Comyn at the bank-led event in Sydney.
The pair discussed how these days almost anyone can use AI to build software and coding mechanisms and how there needs to be a “reframing” of how work will be done in the future.
“The value (of AI) is created in a country where it is broadly infused across that country,” Mr Comyn told the audience of leading business people and bank clients.
“One of the most important areas, both economically and socially, is how do we prepare our citizens, our employees, with the skills and capability for the future?”
Mr Nadella said AI adopters have to go in with “high ambition”, starting with curiosity and willingness to learn.
“But teams and people who are willing to question the status quo are really able to see the productivity gains,” he said.
“I’d say reframing of the jobs to be done is so critical to all of us.”

However, cyber security remains a key issue given the attack on Australia’s systems five months ago, and especially for businesses developing and managing their own code bases.
“Attacks are coming, driven by essentially someone sending a ‘telegram’ message to a harness that’s using open weights – an attack,” Mr Nadella said.
“Nation-state attacks can be done by anybody today … models are getting more and more powerful.
“So AI safety, let’s put it and say, a lot more work needs to be done.”
So while agents have the intelligence layer, the trust layer has to be kept separate to ensure “agent armies” are also protecting networks.
“Our architectural approach is to sort of really have observability at every layer, starting with the model,” Mr Nadella said.
“We need to treat some of these models as insider risk and surround these with very capable layers … and … you also have to manage the outside risk, which is people attacking you.”
Last week, Qantas told its annual general meeting that in 2025 it was fending off 2.3 million cyber attacks a month.
The rate has now risen to two billion attempts on its secure airline systems each month.
Commonwealth Bank on Wednesday announced it was rolling out CodeCheck, a free cybersecurity tool and AI harness to help businesses uncover vulnerabilities in their software systems.
It comes after OpenAI earlier in 2026 began allowing large, vetted global corporations access to its frontier cybersecurity models to fix security weaknesses.

The Australian Cyber Security Centre warned organisations of potential malicious activity more than 4400 times in the last financial year, a rise of 80 per cent on the previous year.
Mr Nadella’s fireside chat with Mr Comyn was recorded on November 2, so the issue of last week’s Microsoft meltdown didn’t come up.
Last Friday, users across the world suffered a massive outage that disrupted services, including Outlook and Teams, caused by a server issue.
AAP