In the World of AI: Week of 25 August 2026, Treasury Warns Australia's AI Use Is Too Shallow

Treasury has told the Treasurer that Australian businesses use AI widely but shallowly, Anthropic has let Claude act on logged-in websites for every paid plan, and Salesforce has joined forces with Anthropic on a sales plugin. Here is the week to 1 September 2026, with one step for each story.

1. Treasury says Australia's AI use is widespread but shallow

ABC News reported on 31 August that Treasury's advice to Treasurer Jim Chalmers warns Australia will miss AI's economic benefits unless businesses move beyond basic use and change how they operate. About two-thirds of Australian businesses report some AI adoption, but fewer than 10% report significant adoption. Treasury said the job losses some people fear have not yet appeared, and that officials are watching young and entry-level workers overseas for early signs. It regards its long-run productivity growth assumption of 1.2% as achievable, with 1.5% to 2% a plausible upside and 0.5% to 0.8% a realistic downside.

For small firms, Enterprize Tasmania's Brian Collins described a "confidence gap more than it is a skills gap", with many owners assuming AI is only for big companies or is too expensive. Treasury said gains depend on investment in processes, business models, management practices and workforce skills as well as access to the technology.

Why it matters: Having a chatbot subscription counts as basic use. The benefit Treasury describes comes from changing how a process runs.

This week: Pick one process, such as quote follow-up or invoice chasing, and write down every step and who does it today. That map is the first step to changing the process, and our answer on a good first AI project shows how to choose.

2. Claude in Chrome reaches all paid plans

Anthropic announced on 26 August that Claude in Chrome is generally available on all paid Claude plans. Claude can now take browser actions without asking approval for each one. It can read the current page, type, click links, move between pages and fill out forms using your existing logins. A classifier checks each action against your original request and blocks mismatches, and you can turn off automatic approval in settings. Enterprise administrators can limit it to approved domains. It does not yet run on other Chromium browsers or on mobile.

Anthropic reports that no attacks hidden in web content succeeded against Claude Sonnet 5 or Opus 5 with these safeguards, and that 0.3% succeeded against Fable 5. It adds that "Prompt injection remains a moving target."

Why it matters: An assistant logged in as you can do your admin, and it can also be tricked by text hidden on a web page into doing something you did not ask for.

This week: If anyone on your team tries browser agents, limit them to low-risk sites first and keep manual approval on for payments, customer records and email. Our guide to human review of AI output sets out where to put the checks.

3. Salesforce and Anthropic launch Claudeforce

Salesforce announced Claudeforce on 26 August, an expanded partnership with Anthropic. It has three parts. A plugin called Salesforce in Claude has 37 prebuilt sales skills, such as meeting prep, deal health review and pipeline review, and can take governed actions on live revenue data. Claude is also a reasoning model inside Salesforce's Agentforce, and the default model for Slack.

Salesforce in Claude is available to select pilot customers now, with an open beta expected in September 2026 and more prebuilt skills from late 2026. The release does not state pricing and says availability may vary by region and customer agreement.

Why it matters: Sales tools are becoming places where an AI assistant can read and update your records, not just draft text. For a business on Salesforce, this is a direction of travel, not a ready-made product yet.

This week: If you use Salesforce, ask your account contact whether you can join the pilot or beta, and what the cost would be. If you use a different CRM, ask your vendor what AI actions on your records it permits.

4. Anthropic releases Claude Fable 5.1

Anthropic announced Claude Fable 5.1, which DataNorth dates to 1 September, as generally available on the Claude API, Amazon Web Services, Google Cloud and Microsoft Azure. Anthropic estimates it costs about 25% less than Fable 5 on typical workloads, and up to about 45% less on highly agentic ones. The main change is a 75% cut in cache-read pricing, to US$0.25 per million tokens, while other prices stay at US$10 for input and US$50 for output per million tokens. Those savings are Anthropic's estimates from four weeks of August usage.

Anthropic also released a companion, Mythos 5.1, with fewer restrictions for approved cyber and life sciences users. It is currently open only to US organisations. A planned service called Enterprise Frontier Safeguards, which would keep prompts and outputs in a customer's own cloud, rolls out in phases from this fall.

Why it matters: Cheaper running costs for tools built on this model could reach you through lower prices or better margins for your supplier. That is only likely if your supplier uses Fable.

This week: Ask any supplier who built an AI tool for you whether a cheaper model would change your costs, and whether they will pass it on.

What to do this week

• Map one process step by step and note who does each part today.
• Limit browser agents to low-risk sites and keep manual approval on for payments and customer data.
• If you use Salesforce, ask about the Claudeforce pilot and its cost.
• Ask your AI suppliers whether cheaper models change your bill.

The Treasury figures are a snapshot reported in the media, and the plan for your own business should rest on your own numbers. If you would like help working out where AI fits, see our AI strategy service and our answer on how to measure AI return.

Sources

• ABC News, "Australia risks missing out on AI economic boom due to 'slow' business uptake, Treasury says", 31 August 2026 (updated 1 September): abc.net.au
• Anthropic, "Claude in Chrome is generally available", 26 August 2026: claude.com
• Salesforce, "Salesforce and Anthropic announce Claudeforce", 26 August 2026: salesforce.com
• Anthropic, "Claude Fable 5.1 and Mythos 5.1", September 2026: anthropic.com
• DataNorth, "Anthropic launches Claude Fable 5.1 with 75% cheaper cache reads", published 2 September 2026, giving a 1 September launch date: datanorth.ai