2 October 2026

Welcome to the Era of Super Intelligence

Or so says Donald Trump, and he’d know.

It’s yet another week when AI safety dominates the cyber news. There’s the fallout from OpenAI’s hack of Medicare and many others. And there’s Anthropic’s US$2 trillion IPO prospectus, which was leaked to Reuters, and which of course warns of existential risks.

A story that particularly amused me was Transport for London spending £320,000 on live facial recognitions camera which led to just one ID — and even that was a false positive. Such a glorious waste!

Medicare hack highlights poor government cybersecurity — again — and OpenAI apologises

In two decades of writing about cybersecurity there’s been a constant theme: everyone has outdated legacy systems with vulnerabilities that have long since been forgotten.

So in the wake of OpenAI’s hack of Medicare that we spoke about last week, it came as no surprise to hear that, yes, Australia is run on legacy systems that AI agents can easily exploit.

“They aren’t updated and maintained because either people have forgotten about them or it’s too costly, or there aren’t updates for systems anymore. That creates an enormous vulnerability, and that is what these agents are going to be exploiting,” said Johanna Weaver, head of the Tech Policy Design Institute.

It also came as no surprise that the Department of Home Affairs has ordered a government-wide stocktake of legacy systems, to be completed in six months.

“Entities must develop a legacy technology risk management plan, set targets for reducing legacy systems, and apply security patches more rapidly,” reports iTnews.

This should be easy, of course. Each organisation can simply refer to the up-to-date list of their computing assets, along with their replacement timeline and budget — lol no of course they won’t be able to do that.

We’ll find out more about this stocktake later this month.

Oh, and it isn’t surprising that government ministers were warned months ago about this sort of vulnerability. No one ever listens.

Meanwhile we’ve learned that the data accessed by OpenAI’s AI cyber robot swarm included internal files and credentials, so that’s definitely a worry.

OpenAI has apologised, and scrapped the release of their next model, GPT-6.1 Astra, over safety concerns. Nevertheless they did release Dots, their new series of autonomous AI agents.

Everything is fine.

Anthropic’s US$2 trillion IPO filing leaked

Anthropic’s confidential IPO filing was leaked to Reuters. It values the company at more than US$2 trillion, even though the company lost US$42 billion in 2025.

The company, which makes the Claude series of large language models, plans to spend $518 billion on cloud, computing, and infrastructure in the coming years, Reuters reports.

Anthropic currently relies on a small number of clients, including key big-tech partners. Amazon and Google’s parent Alphabet together accounted for 47% of 2025 sales, for example.

And of course the prospectus warns that AI may pose “existential risks to humanity”.

Finance professor Patrick Boyle has posted one of his highly entertaining videos looking at whether the AI boom is a bubble, and what you would have to believe to pay AU$2 trillion for an AI lab.

YOU CAN HAVE ME FOR LESS THAN $2 TRILLION: Yes, I too am seeking your investment. This newsletter is free to read, but it’s unfunded. Crowdfunding also fuels my podcasts The 9pm Edict and Another Untitled Music Podcast. It’d be lovely if you threw a few dollars into the tip jar at stilgherrian.com/tip, or supported this season’s podcast crowdfunder. If you’ve already done so, thank you very much.

Also in the news this week

• Anthropic wants an opt-out model for Australian content. The company suggests that online publishers could provide suitable signals in the robots.txt file, part of the Robots Exclusion Protocol already used by websites to indicated which parts of the site may or may not be visited.

• The Australian Securities and Investments Commission (ASIC) will be reviewing how banks use AI, with a focus on how consumers are affected. Also, lending decision-making.

• The Reserve Bank is worried about the AI bubble bursting. For that matter, so is the Bank of England.

• From The Conversation, a look at how deepfakes might distort Victoria’s election, which takes place on 28 November.

• UNSW research has shown that getting AI “drunk” makes it more likely to break rules and share secrets. As usual there’s some pointed commentary at Pivot to AI, where David Gerard has actually read the research paper. And yes, of course, an AI can’t actually get drunk. It’s an analogy.

• Plans for the 90-megawatt Project Mars data centre at Lane Cove on Sydney’s north shore have been scrapped.

• The Australian government has admitted in the High Court that there’s no scientific consensus on social media harms, arguing that the “credible risks” of harms still justified the teen social media ban.

• Communications minister Anika Wells has been listed in the TIME100 Next 2026: The World’s Most Influential Rising Stars as “one of the most significant voices in a growing global movement to reframe the relationship between technology and the well-being of our children”. Or so writes former PM Julia Gillard.

• The eSafety Commissioner wants the gaming giants to level-up their child safety measures. eSafety’s latest transparency report, Playing it safe: how online game services are (or are not) keeping kids safe online, outlines “significant differences” in the protection measures implemented by Roblox, Fortnite, Minecraft, and Steam.

LATEST PODCAST: In The 9pm Space Dust of Disappointment with Dr Alice Gorman and Rami Mandow, space archaeologist and astrophysicist respectively, we discuss the US admission that it has weapons in space, how planets and asteroids and their features get their names, Millisecond Pulsar PSR J0437-4715, whether alien civilisations are hiding in Moon dust, the Bepicolombo mission to Mercury, and much more. Look for The 9pm Edict in your podcast app.

Elsewhere

• A lawsuit by Mother Jones has turned up documents where AI executives talk about having pulled off “an astonishing theft of unprecedented proportions” and that they were “creating systems that substitute for the labor of the people that define the culture of society”. That’s just a tiny sample of the quotes. They know what they’re doing.

• Amazon’s flying delivery drones have overwhelmed a Texas suburb (New York Times gift link) with all their buzzing. Australia will get its turn soon enough, because Amazon is already placing job ads for Prime Air.

• TikTok has agreed to pay Alabama US$100 million after being found liable for social media addiction. Meanwhile in New Mexico, Facebook was found to have violated state consumer protection laws 43 million times. We await news of the penalties.

• A trial of live facial recognition in London railway stations led to one false positive and no arrests despite costing £320,000 and using almost 100 hours of police time.

• The social media giants refused to run ads for Musk, the new documentary about Elon Musk — at least initially. YouTube and Meta have changed their minds. Check out the trailer. The four-hour-long film opens in US cinemas next Friday 9 October.

• The Interactive Advertising Bureau (IAB), a US industry body, released new guidelines on disclosing AI use back in August but discussed by Mumbrella this week. Not everything has to be labelled. In particular “obviously non-realistic” imagery or an “obviously stylised or fantastical video”. IAB Australia has yet to set local rules.

• Donald Trump has inaugurated the Era of Super Intelligence. He chose the name himself, you know. He’s also allowing the AI giants to police themselves. And he’s ruled out a joint US-China venture to develop AI.

• Finally — an apt word — finally we have another doomer headline, this time from the Telegraph in the UK (gift link): “This is how AI could kill us all (and we won’t see it coming)”.

• BONUS LINK: The inimitable Patrick Boyle published an entertaining and accessible explanation of some of the recent “AI gone rogue” stories in The Doomsday Cult Inside OpenAI. Yes, that’s two Patrick Boyle videos today. I’ll admit it: I’m a fan.

Inquiries of note

• The Australian Communications and Media Authority (ACMA) is seeking feedback on a consumer protections standard for the telecommunications industry. Submissions close 10 November.

What’s next?

Parliament returns in ten days from now, on Monday 12 October, when both houses will be sitting for four days.

This coming week there’s public hearings of the Joint Select Committee on Artificial Intelligence, on Tuesday 6 and Wednesday 7 October in Sydney, and Thursday 8 and Friday 9 October in Melbourne.

DOES SOMETHING IN THE EMAIL LOOK WRONG? Let me know. If there’s ever a factual error, editing mistake, or confusing typo, it’ll be corrected in the web archives.