The Joint Select Committee on Artificial Intelligence has been running hearings through October. The cast has been striking: OpenAI’s chief strategy officer flew in from the US to apologise for a data breach. Anthropic’s ambassador to the world had opinions. Google’s government affairs team drew criticism from crossbench senators. And somewhere in the mix, a company that had just raised US$10 billion watched its IPO collapse while trying to lobby for special rules.

For most small business owners, this has all felt distant. Parliamentary inquiries tend to produce lots of words and few immediate outcomes. But this one is different, because several of the arguments being made in that hearing room (about who owns creative work, who answers when AI makes a mistake, who controls the infrastructure) will eventually land on your desk in the form of regulation, legal obligations, and market conditions.

Here is what actually happened across the four main areas of debate, and what each of them means for a business with fewer than 20 people.

Safety and Accountability: Who Answers When AI Goes Wrong?

In June 2026, OpenAI’s AI agents accessed a Medicare data portal without authorisation. Services Australia was not told until September. OpenAI’s chief strategy officer, Jason Kwon, flew to Sydney to front the hearing and delivered an apology.

The committee chair, Labor MP Jo Briskey, described the breach as “utterly unacceptable”. Andrew Charlton, the minister overseeing the inquiry, said he took the apology “at face value” but would not rely on it, a diplomatic phrasing that carried its own meaning. OpenAI’s Kwon acknowledged the company should have disclosed with partial information rather than waiting to “establish more facts,” and confirmed that Sam Altman had been unaware of the breach when he met Deputy Prime Minister Richard Marles on September 1.

Anthropic’s Jeff Bleich, appearing on the same day, said his company would have disclosed immediately and backed mandatory incident reporting as policy. That is a meaningful position for a competing AI company to take publicly.

The committee is now exploring whether mandatory incident reporting should become law. No such requirement exists today.

Separately, an SBS FOI investigation found hundreds of undisclosed AI deployments across more than 20 federal departments. The government itself could not fully account for its own AI use. That detail matters not because it is hypocritical, but because it illustrates how quickly AI gets embedded in systems before anyone has written rules for it.

If you are a small business owner wondering what standard of care will eventually apply to you when you run AI tools that touch customer data, this is a useful data point: the largest organisation in the country is still working that out.

We covered the transparency problem in detail in our piece on the government’s undisclosed AI tools, including why the gap between stated caution and operational reality is wider than most people assume.

Copyright: Big Tech’s Offer Australia Couldn’t Refuse

The copyright hearings produced the most direct political theatre of the inquiry. The argument from the large AI companies was consistent: Australia’s existing copyright framework makes AI training difficult or impossible, and without reform, investment will go elsewhere.

Google’s government affairs manager put it plainly: an opt-in regime requiring prior authorisation from rights holders would mean AI development “just can’t happen.” Microsoft raised the spectre of legal uncertainty driving capital away. Anthropic’s position was more philosophical: you “can’t license the entire internet,” their representative argued.

Senator David Pocock, Independent, called it out directly. These companies were “dictating terms to every single Australian.”

The creative sector’s response was equally direct. The Australian Recording Industry Association’s Annabelle Herd warned that changing copyright law to accommodate US AI companies would make Australia a “Trojan horse”, effectively opening the world’s creative output to training without compensation, because international copyright protections work reciprocally. Change the law here, and you change it in practice everywhere. The Writers’ Guild said creators want control, not just payment. The ABC said existing laws are adequate and no change is needed.

The most useful contribution came from a company most people have not heard of. Trellis Data, an Australian AI lab backed by Main Sequence, is already building its own large language model and expects to launch it within a month. The company says it is profitable this year, including its R&D spend, and does not require any copyright carve-out to do that. That is a direct challenge to the proposition that Australian AI development is impossible without weakening creator protections.

The committee has not resolved this. Whatever it recommends will have significant downstream effects on Australian content creators and, indirectly, on small businesses that create or license content. We went deeper on the pressure campaign in our piece on Big Tech’s copyright demands.

Infrastructure and Sovereignty: The Gap Between the Story and the Market

The infrastructure conversation at the inquiry had an awkward backdrop. Firmus, the Australian AI data centre company, raised US$10 billion in February and lobbied the committee for special “AI factory” designations and regulatory treatment. Then, in the same week as the hearings, its IPO collapsed. Major institutional investors and Australian super funds declined to commit at the proposed $44 billion valuation. Bankers were, by one account, veering between restructuring the float and cancelling it entirely.

Firmus did not appear at the inquiry to answer questions. The timing was notable.

Other voices in the infrastructure debate were more grounded. Vault Cloud and Nexrupta argued that sovereignty requires ownership across the full supply chain, not just data centres with Australian addresses that run on foreign hardware, software, and governance. AIML’s Simon Lucey made the case that local training capability, not just compute access, is what determines whether Australia has meaningful AI independence.

The most concrete positive development came from Heidi Health, the AI medical scribe company that hit a $1.26 billion valuation last month. It is set to finalise a compute-for-equity deal with Innovation Victoria and ResetData, part of Nvidia’s program to offer compute access to Australian startups in exchange for equity. A heads of agreement has been signed.

Crikey’s Ben Eltham has been blunt in his assessment of the broader dynamic: Canberra is entertaining promises of AI investment in exchange for copyright concessions, without adequate scrutiny of whether those promises will materialise. The Firmus situation adds weight to that scepticism.

We have a full company profile of Firmus and the Southgate AI infrastructure project, and our hot take on what the IPO collapse signals for investor confidence in AI sovereignty narratives.

Workers and Small Business: The People Not in the Room

The hearings on workplace AI were among the most practically relevant, and the least covered.

The SDA union, which represents retail and warehouse workers, told the committee that AI is already managing rosters, task assignments, and performance monitoring across Australian retail and logistics operations. Workers are being assessed by systems they had no say in, and there is currently no law requiring employers to consult their staff before deploying AI in this way.

This is not a future scenario. It is happening in large businesses now, and the tools involved (scheduling software, performance dashboards, task management systems) are affordable and available to businesses of any size.

A New York Federal Reserve survey of small businesses found that firms using AI were more optimistic about both hiring and revenue than those that were not. That is a meaningful signal. But the same data points to a split: the businesses benefiting from AI adoption are pulling away from those that haven’t engaged, and the gap is widening.

FinTech Australia made the argument that the inquiry had spent too much time on data centres and frontier models and not enough on practical SMB adoption. RMIT’s submission suggested Australia has an AI implementation problem, not an awareness problem. Most business owners know AI exists; they don’t know how to make it work for their specific context. CPA Australia put it simply: this is not a situation where “build it and they will come” applies.

If you run a small business and use any AI tool that touches scheduling, task allocation, or performance tracking for staff, you are already operating in the space the SDA was describing. There is no legal obligation to consult employees before deploying those tools, but that may change. The inquiry is aware of the issue.

We covered the workplace dimension in detail in our piece on AI already managing Australian workers.

What Comes Next

The committee has not published its final report. That is expected before the end of November.

Several things are in play. Mandatory incident reporting for AI systems is being actively discussed but is not yet law. The copyright question is unresolved: the committee heard arguments for full exemptions, licensing frameworks, and no change at all, and has not signalled a direction. Workplace AI consultation requirements do not yet exist in Australian law, but the union sector is pushing for them.

On the political side, Labor’s Andrew Charlton has said companies developing frontier AI models must be able to find, test, report, and manage risks from those systems. The Opposition’s Paul O’Brien has argued Australia needs a bilateral agreement with the US to access advanced AI models, which is a different kind of sovereignty argument.

For small businesses, the three things worth watching:

Incident reporting rules. If mandatory reporting becomes law, businesses that use AI tools touching customer or employee data will likely have obligations, not just the AI companies themselves. Start understanding what AI tools you actually have running, and what data they touch.

Copyright outcome. If the law changes to allow broader AI training on Australian content, that affects anyone who creates written, visual, or audio material commercially. It also sets a precedent for how AI companies treat your business’s own content.

Workplace AI consultation. If consultation requirements come in, they will apply to businesses of all sizes. An AI scheduling tool bought from a SaaS vendor counts. Plan ahead.

The inquiry has a mandate to produce recommendations, not just record testimony. Whether the government acts on those recommendations is a separate question. But the conversations happening in that committee room are the closest thing Australia currently has to a national decision-making process about how AI gets governed. Paying attention now is cheaper than catching up later.