Australian firms scale AI only as trust builds, Cisco says

Cisco has published research suggesting trust determines how far Australian businesses will let artificial intelligence operate. The study draws on consultations with 25 organisations, including AustralianSuper, Flight Centre Travel Group, Culture Amp and Funlab.

The findings suggest the companies using AI most extensively are not waiting for further advances in the technology before expanding its role. Instead, they are focused on proving systems work reliably, protecting data and keeping people responsible for final decisions.

That approach comes as Australia remains relatively cautious about AI. Cisco cited separate research from the University of Melbourne and KPMG showing that while half of Australians use AI regularly, only 30 per cent believe its benefits outweigh its risks.

Evidence first

The report argues internal confidence grows through measurable results rather than management direction alone. Respondents said trust and AI adoption reinforce each other over time, with successful use encouraging wider acceptance and broader deployment.

"Training, usage, and successful outcomes start to build a flywheel of trust. By training your people, getting them to use AI, and seeing good outcomes, you create the confidence for them to keep going," said Chris Locke, Chief Technology Officer, Flight Centre Travel Group.

At Culture Amp, managers tracked both productivity and accuracy when staff used AI tools. That gave the company data to support broader uptake.

"We could see that the people using the tools were three times faster, while error rates were exactly the same. That gave us statistical evidence to back up the benefits, and that built trust around the adoption of AI," said Paul Keen, Vice President of Engineering, Culture Amp.

The study also highlights the role of peer networks inside companies. Funlab, for example, has set up an "AI Super Squad" with representatives from across the business who meet every four weeks to compare experiences, discuss concerns and share lessons with colleagues.

Measured trust

A central theme in the research is that the most advanced users are not seeking unrestricted faith in AI systems. Instead, they are trying to build what the report describes as calibrated trust, where the level of reliance matches the risk attached to a task.

That means people are expected to know when an AI-generated answer is sufficient, when it needs checking and when a person should take over. In practice, this shifts the focus away from simple adoption metrics such as log-ins and towards signs that users are applying judgement.

"Blindly trusting these systems and their outputs is a huge risk. Trust has to include knowing when to challenge, when to check and when judgement still needs to sit with a person," said Sarah Carney, Head of Artificial Intelligence, AustralianSuper.

One practical measure, according to Cisco, is whether employees override or challenge AI outputs when necessary. Another is whether staff use approved tools instead of unsanctioned alternatives, which can introduce governance and data risks.

Autonomy gap

The research found readiness is not keeping pace with ambitions for more autonomous AI. Fewer than half of respondents said they were very or extremely confident their foundations were ready for the next stage of adoption.

As AI systems move from producing drafts and recommendations to taking actions on behalf of users, trust becomes more directly tied to control. The question for businesses is no longer only whether an output is correct, but whether a system will remain within defined limits.

Respondents identified privacy and security breaches as the biggest threats to trust in AI. Legacy infrastructure also ranked among the top five risks, and more than half said they needed to upgrade infrastructure to support newer forms of AI.

"Trust determines the autonomy you are prepared to give AI. The aim is bounded autonomy, where the agent can act on your behalf but cannot go beyond the boundaries you have established," said Devesh Maheshwari, Chief Technology Officer, Lendi Group.

Cisco linked that concern to a wider control shortfall in Australian organisations, citing separate findings from its Relentless Defence Report showing only 36 per cent of Australian respondents apply consistent controls to AI agents and non-human identities.

For Cisco, the implication is that businesses need stronger technical controls before granting systems more independence. The report recommends monitoring and governance arrangements that can operate at machine speed as AI agents take on more tasks.

Stefan Leitl, Vice President and General Manager, Cisco Australia and New Zealand, said the organisations in the study start with use cases where value can be demonstrated and safeguards made visible before widening deployment.

"Trust isn't a hurdle that the organisations in the Cisco study clear once. What we've learned from Australia's AI leaders in the study is that trust is an essential foundation for the change AI will bring. They start where AI's value can be proven, make their safeguards visible and expand as the evidence builds. That's how they're turning AI investment into real returns. It's a playbook that travels well beyond Australia, for any business looking to build new services and business models on AI," said Leitl.

Carl Solder, Chief Technology Officer, Cisco Australia and New Zealand, said: "The leaders in our research aren't scaling AI first. They're scaling trust. That means data they can stand behind, clear boundaries for every agent and people who know when to step in. Built into the network and security underneath, control stops being the brake and becomes the permission to move faster."