The short version
The most useful AI news in the last week of October was not a new model. It was a run of announcements about who controls AI tools once they can act, what they cost to run, and who gets a say in the rules. For an Australian professional or small business, three things changed: there is now a government checklist for AI agents, the major vendors are exposing more of the controls that make agents governable, and two windows opened to shape how AI is regulated and trained.
Two model releases in the same week, OpenAI's GPT-6.1 Terra and Google's Gemini 4 Carbon, are covered in the November Brief and are not repeated here. The editions for 26 to 28 October and 29 to 31 October carry the detail on each item below.
What changed: agents now come with a checklist
On 26 October the National AI Centre published Understanding and using agentic AI, a free guide for small and medium-sized enterprises. Its four checks before switching an agent on are access, people, visibility and stop: limit what the agent can reach, decide when a person must approve or halt an action, make sure its actions are recorded, and know how to switch it off and recover. It also says human review is not always enough, because an agent can act faster or at greater scale than a reviewer can follow.
The same week, two vendors shipped controls that map onto those checks. GitHub made Copilot Fusion generally available on 28 October. It chooses which models handle each coding task, but only from models enabled in the organisation's model policy, and administrators can remove a model from Fusion without removing it for everyone. That is the access check, applied to model choice. Microsoft's finance announcement on 29 October added accounting skills to Copilot in Excel and Cowork for its Frontier early-access program, with an optional audit trail that records in the chat history what Copilot changed and why. That is the visibility check, applied to a ledger.
Why it matters now: until recently, the governance conversation about AI tools was mostly about which model to approve. This week's material moves it to what the tool can reach, what it records and who can stop it. Those are questions a compliance, audit or claims team already knows how to ask of any system. The guide gives small businesses the same vocabulary without needing a specialist.
What changed: the economics moved, and not in one direction
Several announcements changed what AI work costs, and they pull in different directions.
- OpenAI began offering API credits with eligible ChatGPT Pro and Business plans on 28 October, from US$20 per Business Standard seat up to US$500 a month, expiring each billing cycle and subject to cohort and country. Australian availability is unconfirmed.
- Its release notes for 29 October doubled the Work and Codex allowance on the Pro 100 and Pro 500 plans and cancelled a planned reduction on Pro 200.
- GitHub bills Fusion per model used, at standard rates, with a limited-time 30% discount on paid plans.
- Artificial Analysis rebuilt its Intelligence Index on 27 October so that 60% rests on private or held-out tests, and found cost per index task varying more than 13 times among the ten highest-scoring models it showed.
Why it matters now: a business case written in September may already be out of date. Allowances, bundled credits and routing discounts change the cost of a task without changing the headline subscription price, and some of these offers are labelled limited-time or subject to change. The more reliable figure is cost per completed task on your own work, which only a trial can give.
What changed: two windows opened to shape the rules
South Australia's Royal Commission into Artificial Intelligence opened submissions on its Scoping Paper on 29 October. They close on 30 November 2026 at 11:59 pm, and the call is open to other Australians and international respondents as well as South Australians. The paper covers six themes, from AI in society to safe and trusted AI, and accepts written answers, uploaded documents or voicemail.
On 30 October OpenAI said it will honour a shared robots.txt signal that lets a website state its preference about generative AI training once, for every participating provider. It aims to implement it "in the coming months"; the signal is set out in an Internet-Draft, which is a working document rather than a standard.
Why it matters now: both are early-stage. The Commission's first phase sets the themes later rounds will examine, and the training signal is a draft that providers may or may not adopt. Practitioners who comment now shape the questions, not just the answers.
Dates to watch
Three dates from earlier announcements now sit close enough to affect November plans:
- 12 November: Anthropic's updated Usage Policy, announced on 8 October, takes effect. Where Claude is used in ways that can affect someone's health, legal rights, finances, livelihood or access to essential services, it requires a qualified person with authority to review and change the recommendation, and it requires the affected person to be told AI was used.
- 30 November: submissions to the South Australian Royal Commission close at 11:59 pm.
- 1 December: Google's Gemini API pricing page lists Gemini 4 Argon's paid-tier price at US$2.00 per million input tokens and US$10.00 per million output tokens through 30 November 2026, doubling to US$4.00 and US$20.00 from 1 December.
Three things to test before next week
- Run the four checks on one tool. Pick one tool that already acts without asking, such as an inbox rule, a booking system or an AI feature that sends or updates records. Write one line each for access, people, visibility and stop. A blank line is a finding.
- Price one task, not one plan. Take one repeated task and record what it costs on your current plan this week, including any credits or allowances that changed. Note any offer marked limited-time, and set a reminder to recheck it.
- Decide two owners. Name who in your organisation owns the website's robots.txt decision on AI training, and whether anyone will make a submission to the South Australian Royal Commission before 30 November. If the answer to either is nobody, that is the decision to make.
General information only. Not legal, compliance, financial, or professional advice.
TheAICommand. Intelligence, At Your Command.
