G'day.

Today is about labelling. Two big companies decided this week that you should be able to tell when a machine made something, which sounds like policy until you realise it covers the quote you had Claude tidy up.

In today’s briefing:

🏢Australian AI spending is running ahead of the basics

💳NBN Co says AI isn't inflating its software bill

✍️Claude now hides a watermark in everything it writes

🎵 Spotify will badge AI artists next month

🎣 North Korea's hackers now run AI in-house

New SAP research, run by Oxford Economics across 2,600 businesses including 200 here and reported on 7 August, found local organisations doing more with AI and planning more again. Most of that spending happens one project at a time, so nobody's budget covers the boring groundwork — data quality, governance, and who is accountable when it goes wrong.

Why it matters for you: In a small business that groundwork is one page, not a program. Write down every AI tool in use, what customer data it touches, and whose name is next to it.

Source: iTnews

Its 2026 results reported a six percent rise in other operating costs, driven mainly by moving more systems to cloud subscriptions. Asked whether AI was materially pushing up technology spend, chief executive Ellie Sweeney said no. Translated: the subscriptions are eating the budget, not the robot.

Why it matters for you: Before you add another AI plan, count what you already pay monthly. Most people I train find two tools they forgot they had, and one already includes the feature they were about to buy.

Source: iTnews

The Daily Rundown: Anthropic confirmed this week that text from newer Claude models carries an invisible watermark woven into the words themselves, and generated images get signed provenance data. It survives copy and paste, applies worldwide, and cannot be switched off. Anthropic is explicit that a mark means Claude processed the text, not that Claude wrote it. The detection tool hasn't shipped yet.

Practical takeaway: Use Claude to proofread your own quote and that document may now carry a mark saying AI touched it, even though you wrote every word. Where a client asks you to declare AI use, declare the honest version yourself.

Source: Claude Help Center, Mashable

The Daily Rundown: Spotify announced an AI Persona badge for artist profiles whose public identity looks AI-generated rather than human. From today artists can declare it themselves; Spotify will also review profiles and apply the badge, with an appeal process. Badged profiles are left out of recommendations by default, and the badge is about the identity, not whether AI helped make the music.

Practical takeaway: Disclose it yourself and you keep control of the wording; wait, and someone else's review team writes it for you. Worth a thought if you publish AI-assisted images or voiceovers.

Source: Spotify Newsroom, Bloomberg

The Daily Rundown: South Korean security firm Genians reported on Monday that the North Korean-linked group Kimsuky had set up software to run AI models on its own servers, wired to a private document store. That lets them draft convincing material without sending anything to a chatbot that might log it. Genians also found AI-made decoy documents dressed up as investment reports. Reuters notes the findings are not independently verified.

Practical takeaway: The era of spotting a scam by its clumsy English is over. Verify any payment detail change by ringing the number you already have on file, not the one in the email.

Source: Reuters

Melbourne fintech Eightcap has put AI into its compliance team, which used to monitor huge volumes of customer calls, emails and chats by hand. It now generates transcripts, summarises conversations, finds duplicate records and flags sentiment for a human to review. Staff still handle the complicated cases; the machine handles the volume.

Why it matters for you: This is the pattern that works, and it isn't glamorous. Give AI the reading and sorting, keep judgement with a person, start with the pile nobody wants to touch.

Source: iTnews

DeepL Write — a browser tool that rewrites what you've already written for clarity, grammar and tone, rather than inventing text from a prompt. Paste in a paragraph, pick a tone, get back cleaner versions of your own sentences. It keeps your voice, which is what a customer email needs.

Verdict: Worth ten minutes on your next awkward email.

Planning a week's work is the job nobody asks AI about. Try this:

Here are this week's jobs: [each one with the suburb, roughly how long it takes, and anything that has to happen first]. I work Monday to Friday, 7am to 3pm. Group them so I'm not crossing town twice in a day, and tell me which two will most likely blow out.

The last clause is the point. Anyone can make a list; the useful part is being told where it breaks.

Spotify badging AI artist profiles and Claude's new watermark are both about the same shift: provenance is becoming visible whether you plan for it or not. AI Graphics for Non-Designers has a lesson built for exactly this moment "Did You Make That With AI?" on spotting and managing the tells before someone else points them out.

And the Eightcap story is the pattern worth copying directly: AI does the reading and sorting, a person keeps the judgment calls. That's the backbone of Your AI Customer Support Team, triage on arrival, AI handles the routine, you stay on the cases that need a human. Both are in the AI Creator & Builder and AI for Business tracks at Vortex Academy.