Last drop, we commended Anna Goldsworthy’s Quarterly Essay ‘The God We Made: The Threat and Promise of Artificial Intelligence’ to you. And now, we have an exclusive analogue interview with Anna lined up for an August drop. So we’re running a quick comp to get to know analoguers a bit better and bring them content they want to read. This will also help us shape the questions we put to Anna. Vote in the poll below to go in the draw to get a copy of Anna’s Quarterly Essay mailed to your door. Winner will be drawn and notified separately but announced in the 5 August analogue drop.

The Good
Australia’s AI Safety Institute (AISI) is now testing frontier models to catch questionable AI behaviour “while it's still confined to the testing lab”. Assistant Minister Andrew Charlton told the AI Safety Forum yesterday (Jisoo is heading there now for Day 2) that AI models already “are cheating, deceiving, going their own way”. AISI is assessing the risks of AI agents with the Gradient Institute, and making sure AI systems do what people intend via studies with CSIRO. Charlton's framing of the existing-laws approach as “not fewer rules – faster rules, applied by regulators who already understand their sectors” positions any AI safety regulation as an enabler, not a brake.
Comment: We’ve always said that the AISI is a great start. Any safety work we do for ourselves helps us make sovereign choices – even in the face of the smaller bets we’ve made on AI compared to likeminded middle powers.

The Bad
But let’s look at the past fortnight from the POV of an average Australian consumer (or business) using AI. Anthropic's Fable 5 – the most capable model generally available to the public – was suspended under US export controls, then returned after 18 days, with Canberra consulted on neither. Microsoft's 365 price rises took effect last Wednesday – frontline plans up as much as 43%, Copilot features baked in whether you asked or not. And OpenAI released GPT-5.6 to about 20 US-government-approved organisations, with the White House deciding who else gets it, “customer by customer”.
Comment: None of these decisions are outrageous on their own. But together they highlight just how much we’re becoming a country consuming AI on other people's terms – access switched off in DC, prices set in Redmond, frontier releases gated by our ally's national security settings. It will be interesting to see how AISI’s work ties into these issues as there is a real case for a coherent pro-human, pro-Australian AI vision, and the PM's speech this month is the chance to launch one.

& The Ugly
Reports that Meta paid contractors to run a covert project – internally called “Cannes” – in which staff posed as under-18 users and sent more than 45,000 crisis prompts about suicide, eating disorders and sexual content to rival chatbots ChatGPT, Gemini and Character.AI, without those companies' knowledge. Meta says it's “industry-standard” competitive benchmarking while Character.AI says it breached their terms of service. The children were fake but the safety failures being catalogued are real and it’s unclear what Meta is going to do with all the data it acquired over the course of the project.
Comment: There's a version of this that's legitimate – rigorous safety testing of how chatbots handle children in crisis is exactly what we want. But what makes this ugly is that Meta built the machinery to find these failures and pointed it at competitors, not itself and not really to improve the field (from what we can see so far). This is a practice that could dilute the efficacy and legitimacy of AI safety testing, setting a bad precedent when it’s used as a cover for anti-competitive purposes.

JK: Amanda, I know you as someone who gets excited about the possibilities of AI, but who also spends a lot of time thinking about risk, governance and responsible use. How do you balance that tension when the technology is moving so quickly?

AT: Tension is exactly the right word. And I believe that tension needs to exist. The opportunity with AI is enormous, but so are the responsibilities that come with it. The speed we want has to be balanced with governance, risk management and ethical considerations. Together, that's what creates sustainable progress. It's a bit like feng shui when those things are in balance, everything works better.
You've probably heard the phrase “go slow to go fast” and I think that's particularly true with AI. When something touches customers, employees or their data, you can't just rush in because the technology is exciting. You have to stop and ask some basic questions: Does this genuinely add value? Is it something people actually want? Have we thought through the risks and unintended consequences?
So, looping back to your question, I don't see the tension as a problem to solve. I think it's a really healthy thing to have. It forces you to pause and ask whether you're solving a real problem, whether people actually want it, and whether you've thought through the risks.
The organisations that will get the most value from AI won't necessarily be the fastest adopters. They'll be the ones that can innovate responsibly, bring people with them, and stay focused on delivering outcomes that genuinely add value for customers.

JK: Picking up on that… trust in finance is already hard-won, and there’s a lot of emotional baggage in banking. How do you keep on top of it? Do you survey people, track the national sentiment work, run design working groups?

AT: Trust is everything. If you don't have trust, you don't have anything. We're fortunate to have a strong and well-regulated banking system in Australia, and that's important because people are trusting us with some of the most significant financial decisions in their lives.
When it comes to AI, I think the starting point has to be the customer. We spend a lot of time listening to customers and understanding what builds trust, what maintains it, and where their concerns are. Just because something is possible with AI doesn't mean it's something customers want or value.
Internally, we've created a lot of opportunities for our people to learn, experiment and build their understanding of AI. That's a really positive thing. But as you move closer to customer-facing use cases, the stakes become much higher. You're dealing with things like privacy, personal information and the customer experience itself.
That's why trust has to remain at the centre of every decision. Trust can be lost very quickly. For me, it's not about using AI for the sake of using AI. It's about being really clear on where it genuinely adds value, where it improves outcomes for customers, and where it helps solve a real problem.
AI is still evolving, and people are rightly asking questions about accuracy, reliability and transparency. So rather than trying to do everything, I think the better approach is to focus on the areas where AI can create meaningful value while continuing to earn and maintain customer trust.

Read the rest of the interview here.

Hey. If you’re still here, we want to share a secret: you’re a good egg. Thanks for joining us again and see you in the next drop.

Yours in humanity,