SOUTH AUSTRALIA · THE SA SIGNAL

12 September 2026 · Strategy & Adoption

Seven per cent of SMEs use AI broadly. Governments are moving faster than most businesses.

Australia does not have an AI awareness problem. It has an implementation problem.

Forty-three per cent of Australian small and medium businesses reported some AI use across the three months to February 2026. Only 7 per cent said they used it broadly across their business.

That gap matters more than another model launch.

It means plenty of businesses have opened ChatGPT, generated some copy or tested a summary. Very few have changed a real operating process, measured the result and decided who carries the risk when the system gets something wrong.

The latest signals are coming from several directions at once. Canberra wants evidence about what is holding adoption back. South Australia is preparing a Royal Commission into AI. A leading model provider has published new cases of criminal misuse. In Modbury, a free workshop is trying to help small businesses use the tools for ordinary work.

These are not separate stories. Together, they show an economy trying to move from casual experimentation to accountable use.

The National AI Centre’s latest published SME tracking found that 43 per cent of businesses reported some AI adoption across December 2025 to February 2026. February alone reached 44 per cent.

The harder number is 7 per cent. That is the share making broad use of AI.

Among businesses that were not adopting it, about 65 per cent cited distrust in AI decision-making or a preference to retain human control. Fifty-four per cent said AI was not relevant to their business. Nineteen per cent said they did not know how to use it.

The survey covers at least 400 SME owners and decision-makers in each monthly wave and weights responses by industry, state and business size. Different businesses may be surveyed from one wave to the next, so small movements should not be treated as a precise trend.

The evidence points to a blunt conclusion: access to an AI tool is common. Operational adoption is not.

Read the National AI Centre’s adoption findings and survey methodology.

A new Joint Select Committee on Artificial Intelligence was appointed on 20 August. Its scope includes productivity, regional benefits, SME and family-business adoption barriers, workforce effects, data sovereignty, copyright, fraud, critical infrastructure and the adequacy of existing laws.

Submissions close on 14 September. The committee is due to report on 30 November.

The deadline is tight, but the question is useful: what is stopping an Australian business from moving beyond isolated AI use?

The answer is unlikely to be “we need more impressive demonstrations.” The adoption data already identifies trust, relevance and practical knowledge as major barriers. Useful evidence would describe a specific process, the attempted change, the cost of implementation, the result and the unresolved risk.

The committee also warns that submitters remain responsible for the accuracy of material produced with AI. That sentence captures the governance challenge in plain language. AI can assist with work. It does not inherit accountability for the work.

Read the committee scope and submission details.

The South Australian Government says its Royal Commission into Artificial Intelligence will begin on 1 October and deliver a final report by 1 July 2027.

The commission is intended to examine the economic and social opportunities and challenges AI will bring to the state. Business, industry bodies, unions, technology developers, academics, independent experts and creative industries are expected to provide evidence.

As of 12 September, the commissioners had not been announced and the terms of reference were still being determined.

South Australia has chosen its most powerful form of public inquiry before publishing the exact questions it wants answered.

For business, this is not yet a new compliance obligation. It is a signal that AI’s effects on work, investment, public services, creative industries and infrastructure are moving into formal state policy.

The strongest future contribution will be evidence, not positioning. Which systems are already being used? What changed after deployment? What did implementation cost? Where did errors occur? Who reviewed the output? Which risks could not be resolved?

Read the Office for AI announcement.

Anthropic’s threat intelligence report published on 10 September describes activity it says it detected and disrupted between December 2025 and August 2026. The cases span cyber operations, influence operations, surveillance, scams and fraud, biological misuse, weapons development and attempts to copy model capabilities.

One operation used several AI instances in separate roles, with one writing code, another researching and another reviewing. In another case, an AI-powered deceptive service reportedly generated roughly 2.36 million messages over two weeks.

This is evidence from a vendor about misuse of its own systems. It is not an independent measure of how common these attacks are, and it does not establish a specific Australian threat rate.

Attackers are not limited to asking a chatbot for suspicious text. They are assembling repeatable workflows, assigning specialised roles and operating at volume.

A business that connects AI to email, documents, customer records or internal systems should treat permissions as part of the risk model. The question is not only what the model can say. It is what the connected system allows it to see and do.

Read Anthropic’s September threat report.

A free beginner-to-intermediate workshop is scheduled for 16 September at Tea Tree Gully Civic Centre in Modbury. The advertised exercises include emails, social media content, business documents, spreadsheet analysis and prompt libraries using ChatGPT and Claude.

This is practical local support, and it addresses the 19 per cent of non-adopting SMEs that say they do not know how to use AI.

It should not be mistaken for evidence of business impact. The event has not happened. Its outcomes have not been measured. The promotional claim that an email can be written in two minutes says nothing about the time required to check it, correct it or deal with a mistake.

Training is an input. A changed process, measured result and controlled risk are evidence.

Check the Modbury workshop details.

Government is asking why adoption remains shallow. South Australia is preparing a major inquiry. Providers are documenting organised misuse. Local programs are teaching entry-level skills.

The missing layer is inside the business: a decision about where AI is allowed to operate, what result would justify its use, what information it may access and who signs off on the output.

The Office of the Australian Information Commissioner recommends that organisations do not enter personal information, particularly sensitive information, into publicly available generative AI tools. It also advises businesses to assess whether a product is suitable for its intended use and how human oversight will work.

That is a stronger starting point than a list of prompts.

Read the OAIC guidance for commercial AI products.

Name the task, the tool, the information it receives, the person responsible for checking the result and the evidence that would prove it is useful.

If those five facts are unclear, the business is still experimenting.

One question to take away:
Is your business using AI, or has someone simply opened an account?

Until next time,
The AI Brief

Practical ideas. Real business. South Australia.

Images generated with ChatGPT. Illustrative editorial scenes.