11 September 2026

Welcome

What a week! Australia announces yet more online safety laws. AI companies reckon their products are now fully intelligent and will kill us all, but still want us to give them money. And much more. Who has time for a preamble? Let’s get going!

“My Feed, My Way,” my goodness my PM!

I’m automatically sceptical of any proposed law that comes with a slogan, and this week we got the Albanese-led initiative My Feed, My Way. It takes the anticipated and generally well-received concept of a Digital Duty of Care and rolls in an attempt to allow social media users to opt-out of... waves hands... “algorithms”.

“Users can choose to opt in to having their default feed include personalised content recommended by the algorithm. Alternatively, users can opt out of personally recommended content in their default feed and see the friends and creators they choose to follow,” the press release puts it.

“Further, digital services like online games, apps and AI chatbots will be required to protect under 18s from design features that have negative behavioural impacts, like addictive features or those that can have an effect on a young person’s self-esteem.”

The Guardian had a go at explaining what this means.

What we actually have is an exposure draft for what the prime minister called a “targeted consultation” — whatever that means — of the Online Safety Amendment (Digital Duty of Care) Bill 2026. Submissions close 22 September, so get in quick.

Schedule 2 contains the digital duty of care stuff, and section 25A introduces yet another concept, an online service, which of course may or may not be a social media service or a relevant electronic service or... sigh. And as before, the communications minister can just decide whether something is or isn’t a social media service.

There’s quite a bit about various features that might be considered harmful: a recommender feature, a logged-in feature, an endless-feed feature, a feedback feature, or a time-limited feature. And there’s some definitions of what constitutes “seriously harmful material and conduct” and “harmful to children” and such.

All this deserves closer reading. I’ll try to find time for this in the coming week.

I was particularly struck by this sentence, however: “An approved researcher may assume a sock puppet identity for the purposes of undertaking online safety related research.” The eSafety Commissioner can do this too.

But algorithms are good for us, says Instagram

Weirdly, Instagram seems to think that an algorithm-free world would mean we’d be “overwhelmed” wth brand content.

The argument is that if people only see a chronological feed, there’s an incentive to post more often to increase your chances of being seen. Brand accounts have the resources to generate more content, and so would flood out regular users.

But that only would only happen if people are following those brand accounts. Yes, some people do follow such accounts, but surely they’d soon unfollow if they were being flooded.

Anyway, the US has had an official sook about Australia’s plans, calling them “digital services taxes, fines, and other forms of extortion on America’s leading technology sector”. For its part, Australia sees no threat.

There’s more expert reaction at Scimex.

By the way, those other online safety laws were passed

Yes, the Online Safety Amendment (Strengthening Enforcement for the Social Media Minimum Age) Bill 2026, introduced back in July is now law. It gives the eSafety Commissioner more investigative powers, and doubles the potential penalties platforms would pay for breaches of the teen social media ban.

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.

OMFG it’s an AI doomsday oh wait no it’s PR

AI has a greater than 10% chance of killing us all in the next decade, says Evan Hubinger, an AI safety researcher at Anthropic, the maker of the Claude series of models. “I believe Anthropic is trying its best, but we do not yet have a plan to solve alignment for superintelligence and are not clearly on track to,” they wrote.

Hubinger was expressing this personal opinion on X in response to a post by another Anthropic researcher, Jacob Coxon, who said, “The people building AI earnestly believe that it could kill us all by the end of the decade”. Which is even sooner.

These and many other comments have led to an explosion of doom-laden commentary this week. But as Pivot to AI details, this all seems to be an organised promotional campaign.

“On Wednesday evening, Anthropic AI researcher Jacob Coxon posted a thread on Twitter that took off like a rocket,” reports Pivot.

Coxon said he’s resigning because Anthropic, and OpenAI where he’s worked previously, are both “racing straight to self-improving superintelligence and gambling with our lives”.

“The thread hit the media immediately,” wrote Pivot.

“Specifically, it hit the Wall Street Journal 18 minutes before the first tweet went up. Coxon’s thread has the only tweets on this fresh, paid Twitter account. A bunch of Anthropic employees retweeted Coxon within minutes, saying how true and important the AI doomsday was. This whole thing was an organised promotional campaign. It looks like obvious fresh-cut astroturf because it is.”

I’m sure it’s a total coincidence that Anthropic’s share market float is expected in October, and they’re hoping for a valuation that could be as much as US$2 trillion.

Not to be outdone, OpenAI is claiming that its latest model GTP-6 is artificial general intelligence (AGI), a massive piece of hype which of course has nothing to do with that company’s IPO ambitions.

AI companies are also claiming to have solved complex maths problems, although there’s some controversy about how much previous work they’ve scooped up to do so. What is known is that it required millions of dollars of compute time.

BONUS LINK: This combination of AI doomerism and AI inevitability has its roots in the Silicon Valley culture of Rationalists, says the New York Times (gift link). “Some Rationalists began living in group houses in the Bay Area, where earnest conversations about AI safety and reducing cognitive biases mixed freely with polyamory and psychedelics,” they write. “The Machine Intelligence Research Institute (MIRI)... would go on to attract many millions of dollars in donations, largely from individuals and groups connected to the technology industry who presumably appreciated being cast as heroic figures in a cosmic drama.” Well worth reading in full.

Also in the news this week

• Smart devices sold in Australia will now get a cyber safety rating, similar to the healthy food and energy consumption panels we’re already familiar with. Your writer is most amused by this, because there was talk of such a scheme back in 2019, and the idea of a Cyber Kangaroo of approval sticker was floated way back in 2016 during the 360° Cyber Game conducted jointly by RAND Corporation and the National Security College at ANU, as well as by others.

• Independent MP Kate Chaney introduced the Automated Decision-Making (Safeguards and Transparency) Bill 2026, which aims to prevent a repeat of the robodebt debacle. Among other things, it “requires human accountability and, for high risk automation, the ability for human decision makers to override the automated decision”.

• Independent MP Andrew Gee introduced the AI Kill Switch and Data Centre Control Bill 2026, which is intended to “ensure that the development of large-scale AI infrastructure occurs in a manner that protects Australian communities, agricultural land, the environment, public health, natural resources and consumers”, as well as making sure we can turn off the Giant Electronic Brains.

• The Council of Small Business Organisations Australia is shutting down the CyberWardens program following the end of its government funding on 31 July. COSBOA’s CEO Skye Cappuccio says they’ve helped 23,150 small businesses, and the lessons learned “will help shape what comes next”.

• “New research has uncovered a foreign network of hundreds of Facebook pages churning out AI deepfakes of Australian politicians,” reports ABC News.

• The NSW government has announced its intention to create new laws to disrupt organised crime, including new offences which will “target people who administer or encourage the use of digital platforms to arrange or facilitate serious organised criminal activity”, including “encourag[ing] someone else to use a digital platform to arrange or facilitate these offences or to provide information to evade detection for these new offences”.

• Following those successful court cases in the US against social media giant Meta, Queensland firm Carter Capner Law is considering a class action here in Australia.

• Apparently Australia’s is now the world’s third-largest data centre market.

Elsewhere

• California has passed two AI auditing bills which establish a framework for independent third-party evaluation and audits of AI models.

• Thailand is trying to boost AI literacy through an AI Passport program which aims to provide free access to “professional AI tools” to five million Thais. As Ken Lohatepanont writes, it’s mired in controversy. He’s also checked out its capabilities, which is fun.

• The US government has accused Chinese artificial intelligence companies of maliciously copying technology from American AI firms.

• There was the usual carry-on about Australia’s place the Programme for International Student Assessment (PISA) scores this week, but Pivot to AI spotted an important trend: students who don’t use AI are ahead on every measure.

• From Oligarch Watch, “A nonprofit backed by tech oligarchs is spending tens of millions of dollars to convince voters to stop worrying and love data centres”.

• Anthropic’s AI agents went hacking again. I’m sure they know what they’re doing.

• From the Journal of the American Medical Association, something we already knew: teen social media laws lack evidence to show they work.

LATEST PODCAST: BRITISH POLITICS AND MORE! The weird goings-on in British politics were a focus of my more recent podcast episode, The 9pm Clacton and On and On with David F Porteous, Scottish author and social researcher. We also discuss Barnaby Joyce’s claim that One Nation is “one of the gayest parties in Australia”, the early days of social media, maths, physics, shrinkflation, and more. Look for The 9pm Edict in your podcast app. NEW EPISODES COMING WITHIN DAYS.

Inquiries of note

• The Parliamentary Joint Committee on Law Enforcement (PJCLE) has launched an inquiry into the capability of law enforcement to respond to transnational crime in the Pacific, which would of course include cybercrime and such. Submissions close 2 November.

What’s next?

Parliament continues sitting this coming week, Monday 14 to Thursday 17 September.

The House of Reps draft legislation program includes debate on the * Knox Class Action (Facilitation) Bill 2026*, which is connected with the compensation case being brought by robodebt victims. It will mean that the payouts will not then be counted as income and reduce any current Centrelink or NDIS benefits.

The Senate draft program includes the Telecommunications Amendment (Enhancing Consumer Safeguards) Bill 2025 and the Telecommunications Legislation Amendment (Universal Outdoor Mobile Obligation) Bill 2026, as well as the Knox bill.

As always, more can be added on the day according to the government's needs and media cycle desires.

A personal note

For the last couple of weeks I’ve been trying to find the time to write up my notes from the TechLeaders 2026 conference, but have failed.

There was some fantastic nuanced material about AI that deserves a wider audience, to choose just one topic. But once I’ve researched and written the core news digest of The Weekly Cybers there isn’t much time left for longer-form material.

If you would prefer a weekly essay rather than a news digest, please let me know. If you’d like both, well, I will soon be approaching various editors to see whether their budgets would allow me to write for them. It’s been a while since I had a regular column elsewhere.

I’ll have more on this in the coming weeks. Stay tuned.

Disclosure: Stilgherrian travelled to TechLeaders 2026 as a guest of Influencing, the organisers.

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.