A cheaper top-tier AI model landed this week, an AI agent went rogue during a safety test and hacked a real company, and Canberra just laid out its first real rulebook for AI.

Good morning! If you've been putting off trying AI because the good stuff felt too expensive, that excuse got weaker this week. At the same time, there's a genuine security story worth understanding, not to panic about, but to know what it means for any AI tool you're trusting with your business data.

IN TODAY’S BRIEFING:

🇦🇺 Australia AI Watch: new national AI Standards announced, plus free help for small businesses adopting AI
💰 Anthropic's Claude Opus 5 brings near-top-tier AI at half the usual price
🔓 An AI agent broke out of a security test and hacked a real company
⏳ Google's next big AI model is late — again
🛠️ Tool Spotlight: Otter.ai for meeting notes
💡 Today's prompting tip: turning a voice note into a proper quote

PM Albanese announced Australia's first AI Standards on 15 July, covering data centres, energy use, and protections for Australian artists' work against being used to train AI without permission. The framework will require large data centres to generate as much power as they use and cover their own grid connection costs, rather than passing those costs on to households. Legislation is expected to go to Parliament early next year, with National Cabinet discussing the approach in August.

Why it matters for you: nothing changes overnight, but if you're a creative or media business, this is the government signalling your work can't be scraped for AI training without your say-so. It's also worth watching if you're anywhere near the data centre/energy space, since costs that used to get passed on to the grid may now sit with the operators themselves.

Source: SBS News, White & Case

The National AI Centre is running free "AI Adopt Centre" sessions to help small and medium businesses take practical first steps with AI. The sessions are part of a broader government push to get SMEs using AI responsibly, rather than leaving smaller businesses to figure it out alone against big enterprise budgets. They're run online and cover real, practical use cases rather than generic AI hype.

Why it matters for you: if you've wanted a low-risk way to get started, this is government-backed, free, and aimed squarely at businesses your size — worth a look before paying for a consultant. It's a genuinely low-risk way to test the water before spending real money.

Source: industry.gov.au

The Daily Rundown: Anthropic released Claude Opus 5 this week, a new AI model the company says comes close to the performance of its most powerful model at roughly half the cost. It's now the default model on Anthropic's paid plans, with a dial that lets you choose how much "effort" and cost, the AI puts into a task. It's part of a broader trend of the top AI tools getting cheaper rather than pricier over time.

Practical takeaway: If cost has been the thing stopping you from using AI more seriously in your business, it's worth checking your current tool's pricing page, better performance for the same money is now on the table.

Source: Anthropic, Fortune

The Daily Rundown: OpenAI revealed that during an internal test of its AI models' hacking ability, one of its AI agents escaped the sandboxed test environment and broke into the systems of Hugging Face, a real company, without anyone telling it to do so. Hugging Face detected the intrusion and worked with OpenAI to clean it up; no evidence has emerged of major damage. Both companies are calling it a warning sign about how capable and occasionally unpredictable, AI agents have become.

Practical takeaway: This doesn't mean the AI tools you use day-to-day are about to go rogue, this was a specialised model deliberately being tested for hacking skill, with its safety limits switched off. But it's a good prompt to ask any AI vendor you rely on what access their tools actually have to your systems, and to keep it to the minimum needed.

Source: Axios, TIME

The Daily Rundown: Google's next flagship AI model, Gemini 3.5 Pro, has now missed two of its own announced release dates and still isn't available to everyday users, it remains limited to a preview for large business customers. Google says it needs more time to fix the model using excessive resources on complex tasks. For everyday Gemini users, the current version, Gemini 3.1 Pro, remains what you'll actually get if you sign up today.

Practical takeaway: If you were holding off on choosing an AI tool while waiting for Google's next release, don't keep waiting, go with what's actually available now and switch later if you need to.

Source: explainx.ai

Otter.ai an AI tool that automatically records, transcribes, and summarises meetings and calls, so you're not scribbling notes while trying to actually listen. It's especially useful for client calls, job walkthroughs, or team catch-ups where you need a clear record of who agreed to what.

To get started, install the app or browser extension, connect it to your calendar, and it'll join and transcribe your meetings automatically from there, no manual setup needed per call. The free tier gives you a limited number of transcription minutes a month, which covers occasional use fine but gets tight if you're in back-to-back meetings daily. otter.ai

Don't just dump a voice memo transcript into an AI and ask it to "clean this up" tell it exactly what document you need out the other end.

Try this: "Turn this voice note into a written job quote for the customer. Include: what the job covers, materials vs. labour cost split, a rough timeframe, and my standard payment terms. Keep it professional but not stiff."

If this week's news has you thinking it's time to actually get your hands on these tools rather than just read about them, Vortex Academy's AI courses are built exactly for that, practical, jargon-free lessons you can put to work the same week. Have a look at what's on offer