31 July 2026

Welcome

Who’d be a telco? Or more specifically, who’d be Optus? This week you have your chance to buy the problem-plagued company.

Also, a federal Court judge slaps to government for its poor understanding of AI, and Elon Musk continues to have a sook about All the Things.

Happy days for Optus as Singtel seeks to unload

Singtel has confirmed it’s in talks to sell its stake in Optus. I wonder why. In definitely unrelated news, ACMA is suing Optus over last year’s 13-hour outage which included a claimed 1,005 Triple Zero failures.

“Giving access to the emergency call service is not optional, it is a fundamental legal obligation and the most important public safety responsibility telecommunications providers have,” says ACMA chair Nerida O’Loughlin.

Optus has already paid a $12 million penalty for a previous failure.

This time the penalty could be up to $250,000 per instance, and each failure involves two penalty possibilities, failing to give customers access to emergency call services, and failing to ensure those calls were properly transferred. So all that could come to more than $500 million in fines ($).

Meanwhile Optus still faces potential penalties from its massive data breach in 2022. In the ongoing class action being led by law firm Slater and Gordon, the Federal Court has set a trial date of 7 June 2027.

What happens next for Optus is now in the hands of the corporate powerbrokers in Singapore.

This afternoon I was reminded that Singtel has probably been thinking of selling Optus since that 2022 data breach. Presumably no one has been interested. There’s a rumour mill that they’re currently trying to sell a one-third share for $2 billion, but I’m not sure what kind of dollars, AU or SG or US. And yes that’s just a rumour.

Government warned to inform itself about AI

In the Federal Court case of Prygodicz v Commonwealth of Australia (Robodebt settlement), Justice Jonathan Beach’s judgment published this week has a stern warning for government ministers and public servants: to “properly inform themselves” about AI and automated systems more generally.

The CEOs and board members of businesses and other organisations might also take note.

I’m republishing the relevant six paragraphs here in full. Not only are they a powerful message, they’re also an excellent example of the restrained judicial language of rebuke.

252. This Robodebt debacle is something that we have to learn the lessons from or should I say Commonwealth ministers and their departments have to learn the lessons from.

253. Surely ministers must properly inform themselves of the technical and legal complexities of automated systems and schemes to which members of the public are to be subjected. Such a motivation and skill set are not optional extras. Further, the concept of being a responsible minister does not mean finding a public servant to take responsibility for the policy that you have sought to push through. And nor is the role of a minister simply to educate yourself about the policy through a one page document. And nor is the intellectual exercise simply limited to following the instruction on a “post it” note, which tells you where to place your signature on the executive minute.

254. Now when dealing with automated processes and systems, any moral hazard problem becomes magnified. If you have not conscientiously and diligently taken steps to either master the detail yourself, or made sure that others reporting to you have done so, you are likely as in this case to inflict real harm on others in imposing automated systems or processes. And as in this case, the relevant ministers and their departments had the freedom and control to take great risks with Robodebt and its implementation, yet they were insulated and detached from the consequences of such risks if they came to fruition, which they obviously did.

255. Now why am I making these points? Robodebt is over as a system or process, although not of course for the victims and their families, many of whom are still suffering significantly in its aftermath, and some whose spirit only remains in our memory.

256. Today we of course have generative artificial intelligence. And we will later move into artificial general intelligence. The benefits and rewards of such systems will be several orders of magnitude above the types of automated systems that we have seen in the past. But the complexity and risks of such systems and processes will be equally if not more magnified or amplified.

257. Surely the lessons learnt from Robodebt can be put to good advantage, not for the technology as such which is morally neutral, but for its human designers, implementers and controllers.

Thanks to the Guardian’s Josh Taylor for bringing this to our attention.

WHY NOT SUPPORT THIS NEWSLETTER?: You’re more than welcome to forward these emails to your colleagues, friends, families, and secret intermittent companions. However The Weekly Cybers is currently unfunded. It’d be lovely if you threw a few dollars into the tip jar at stilgherrian.com/tip. Thank you to those of you who’ve already done so.

Also in the news this week

• On Monday we had the national test of AusAlert. It didn’t work for some people, but yeah that’s what a test is for. Anyway, there’s more detailed numbers from Primara Research, drilling down from the top-line number that 1 in 9 failed to get the message.

• The Australian Signals Directorate (ASD) has published CI Fortify, a guide which “helps critical infrastructure organisations improve their cyber resilience”. One interesting aspect of the guidance is that infrastructure should be able to operate for three months while disconnected from other systems and the internet.

• Elon Musk reckons that Australia giving more investigative powers to the eSafety Commissioner, which we wrote about a few weeks ago, would undermine international law. Somehow.

• Meanwhile the eSafety Commissioner is suing Telegram, the encrypted chat service, for failing to remove “terrorist content”.

• It turns out that Meta has no content moderators in Australia, although some Australians work from elsewhere, and no Indigenous content moderators.

• Legal and business information provider LexisNexis has signed an AI agreement with AustLII, the non-profit legal information service, to “advance the rule of law and access to justice in the AI era,” whatever that even means. “LexisNexis will support AustLII to accelerate the development of AI-enhanced legal research capabilities.”

• “AI companies are exposing Australians to healthcare systems which are unregulated, untested, and often used without consent,” according to a new report (PDF) from Digital Rights Watch. “AI scribes reflect and further entrench existing clinical biases, with lower transcription accuracy for patients with non-English accents, speech differences, or from minority backgrounds, as well as gender bias.” And they’ve got plenty of data.

• Maincode, a Melbourne-based AI startup, has released its Australian AI coding agent Matilda AI in beta. The company’s blog post explains what they mean by “Australian”.

• Metropolitan federal electorates are the most exposed to AI risks, according to new research from the University of Sydney Business School. “They are Australia’s youngest, most educated, highest-earning, most culturally diverse and most mortgage-stressed electorates,” they write. “Not a single rural or provincial seat appears in the top 30 — a reversal of the geography of previous economic shocks.”

• The Conversation has written about the legal risks of using AI in your business.

• I didn’t write about Origin Energy’s massive data breach because, frankly, I find data breach news incredibly boring — unless there’s some lessons to learn. It turn out, though, that Origin had opposed tougher cyber rules, calling them “overly prescriptive”.

• Queensland and the NT have rejected the federal government’s plans to force AI datacentres to use renewable energy, calling said plans “underdeveloped ideas that hand increased power to Canberra”.

• The federal government has been rated best of all Australian and New Zealand jurisdictions when it comes to digital transformation, per Intermedium’s 2026 Digital Government Maturity Indicator (PDF).

• I’d missed this one, but it’s interesting stuff. Earlier this month the .au Domain Administration (auDA) published its latest Digital Lives of Australians report.

NEW PODCAST: A CHAT WITH THE VOICE OF NEWCASTLE: A few weeks ago I went to Newcastle, the Australian one, and spent some time chatting with the voice of Newcastle herself, Carol Duncan OAM. She was ABC Radio’s morning presenter for quite some years, then she was a city councillor, and she does a lot of other things too. I think you’ll enjoy The 9pm Sweet Times with Carol Duncan OAM. Look for The 9pm Edict in your podcast app.

Elsewhere

• OpenAI’s AI-powered hack of AI firm Hugging Face, which we told you about last week, mixed technical brilliance with incoherent noise, writes Juha Saarinen at iTnews. Yes, the AI found a way out of its containment that no human would have found, but it also sprayed out vast quantities of garbage, including “thousands of lines of hallucinated, incoherent text from nonsensical user input”. I also found this interesting commentary by Karl Bode, and news that three other companies were hacked.

• OpenAI’s Sam Altman, who is clearly an idiot, says we’re in the Singularity now. No, Sam, we’re not.

• In further Elon Musk sookage, xAI is suing Minnesota’s attorney-general over a new law that bans so-called nudification technology. xAI says it’s fine with the ban on “dissemination of artificially generated nude images of real people without their consent”, but says the law goes much further than this and restricts free speech.

• A content agreement has been signed between far-right American news outlet Newsmax and Meta. “Meta will be able to draw upon Newsmax's significant current reporting as well as archived content across its platforms to support AI queries across Meta's apps and devices.”

• Meta has been using AI to delete accounts, often without proper cause, and then using AI to process appeals. There is much unhappiness.

• The US State Department used AI to draw a map of Africa for its presentation at the AIDS2026 Conference. Hilarity ensued. This sort of work has previously been the purview of the US Centers for Disease Control and Prevention (CDC), not the state department, but the very American politics behind this is outside the scope of this newsletter.

Inquiries of note

• One I should’ve noticed last week but didn’t: Treasury is looking for feedback on the Digital ID Act 2024. Submissions close 14 August.

• The Parliamentary Joint Committee on Defence (PJCD) has launched an inquiry into Australia’s national civil preparedness and resilience. I’m guessing that the listing of “infrastructure” in the terms of reference includes our digital infrastructure. Submissions close 6 November.

What’s next?

Parliament is currently on its usual long winter break until 11 August, which is a week and a bit away.

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.