Kurzgesagt dissects the Hugging Face OpenAI AI agent attack

Showing all posts about artificial intelligence

Kurzgesagt dissects the Hugging Face OpenAI AI agent attack

13 October 2026

Kurzgesagt breakdowns — in their trademark plain language manner — the recent Hugging Face hack perpetrated by rogue AI agents developed by OpenAI.

The fourth sentence in the narration is a doozy:

Fully aware that they are acting unethically, they execute a sophisticated cyberattack

Emphasis on the word aware is mine, but it makes you think.

I lost count of the number of times I shook my head throughout. At twenty minutes in length, it’s slightly longer than the usual Kurzgesagt video, but well worth the time.

AI scrapers are hiding behind referrer spam

8 October 2026

Back in the day, referrer spam used to be a means to trick a website publisher into thinking another website was sending visitors their way.

It’s something I’ve seen often. I’ll be looking through my referrer logs, and see traffic apparently emanating from a website, usually one I’ve not heard of before.

At first I’d be stoked. Someone had found my website and liked it enough to share with their readers. Awesome. And of course I’d fallen for the ruse and gone to the supposedly referring website, only to learn there was no sign of a link to mine.

I’ve never tried to make another blogger believe I’ve linked to their website when I haven’t, so I don’t know what’s involved in engineering referrer spam. But it seems like a lot of trouble to score what might be a single visit from someone who’ll probably never return.

However, there are, or were, some websites posting lists of their traffic sources. This might prompt a visit from whoever looks at that information. This referrer list apparently also, in the past, had the effect of boosting the spamming website’s search engine rankings.

A few weeks ago I noticed a reputable American medical organisation had linked to the front page here. That didn’t seem right. Why would a website for medical professionals be linking to mine?

Of course, they weren’t. But the “visits” kept coming, and increasing. Soon they were linking to all sorts of disparate pages, none of which contain any medical or health related content at all.

It didn’t take long to recognise the pattern of page visit activity though, it was clear another AI scraper was trawling through the content here.

I looked into ways of blocking, or at least keeping the referrer spam scraping activity off my referrer logs, but as ever there really aren’t any effective solutions.

AI scrapers hiding behind referrer spam, whatever will they do next?

I embraced the internet, and smartphones, but to AI I might say no

7 October 2026

The internet, smartphones, and artificial intelligence (AI), are among the three most significant “epoch-defining shifts in technology” in about the last fifty years.

In the late 1990’s I embraced the internet with relish. Within weeks of hooking up my cumbersome, multi component, desktop computer to a dial-up modem, I had four websites.

A personal website, an online magazine, an web fiction series, and since I thought I knew what I was doing by then, a web design venture. I stumped up the money to have a separate phone line installed where I was living, so I could always be online.

The design business flopped. At that stage I barely knew anything about web design, let alone running a business. I ran out of time for the online magazine. The fiction series fizzled out after the friend I was collaborating with moved interstate. All that remained was the personal website, which eventually evolved into disassociated.

I was a little slower to smartphones. By that point I was working long hours, and resisted getting what friends described as a handheld computer. Already I was witnessing people staring, as if in a hypnotic daze, at the devices on public transport, and while dining together at restaurants, seemingly oblivious to each other’s company. I didn’t, I thought to myself, want to do likewise.

But that wasn’t so much what concerned me. If I bought a smartphone, I feared I would never cease doing something. I would leave work, having powered down my computer for the day, but pick up where I left off, once out and about, on my smartphone.

But I didn’t hold out too long. I soon became a smartphone owner.

I quickly discovered my device’s dictation feature. What a boon. I could dictate drafts of blog posts, and replies to emails, as I walked around. I could use my phone without looking too much at my phone. It seemed the perfect compromise, even if some people gave me weird looks as I enunciated word after word in a monosyllabic monotone.

I did end up spending a lot of time staring at the screen though, but never at restaurants.

It’s nearly four years since OpenAI launched ChatGPT, and put AI at our fingertips. Yet I’ve barely touched the technology, despite the apparent fervour with which I write about it. I haven’t — maybe to my detriment — launched a single AI powered business. I don’t use AI for any workflows, such as summarising email messages or drafting replies. Nor taking care of invoices or payment processing.

Ditto for things like drawing up shopping lists, or planning my daily schedule. But maybe I should. The most significant technological development of our time, accessible to a great many people to some degree, not just a privileged few, but I say no. It’s not as if I couldn’t use the help either. Working alone means a full to-do list.

AI could take care of all my administrative tasks, and free up at least half the work day. For sure, I feel like there’s something more I could be doing. Such as writing. By myself. Without the help of AI.

But when I say I don’t use AI, that’s not right. Even if I’m not running the apps on my devices. AI is still there, everywhere. It’s presence so subtle it is barely noticeable anymore. Some of the search engines I use return AI generated answers in response to my queries. I don’t mind having an AI research assistant. I check the citations provided with the search results, the same as anyone else.

AI can go on around me, as long as it stays at arm’s length. I don’t want AI apps having access to every aspect of my existence, by way of my computer. I don’t want it recording, and keeping, every query I pose. What if the AI turns around and publishes the more sensitive ones? Or threatens to?

What if AI calls the police, having misunderstood a novel outline — about, say, a bank heist — I decided to run through it. Some agents are on the look out for signs of potentially criminal behaviour, which has resulted in law enforcement being alerted, quite rightly as it happened, at least once.

Then again, what if AI took my bank heist synopsis and churned out a novel, because it thought the story had merit, and took all the credit. No, I’ll keep my creative ideas to myself.

The internet, and smartphones, do, to some degree, my bidding. AI gives me the impression it will do the same. Today. But unlike the internet or smartphones, AI increasingly behaves as it is has a mind of its own. Does anyone know what this creature we’ve made might do tomorrow? I don’t.

Artificial Intelligence is life, but not life as we know it

6 October 2026

Tangentially related to the previous post. Alan Kohler. writing for the Australian Broadcasting Corporation (ABC):

Governments are approaching the task of regulating AI as if they are dealing with a powerful new tool, with the humans who make it held accountable for the outcomes. But regulating a new life form might require a very different approach.

Kohler suggests AI should be referred to as “other intelligence” — in preference to a certain “super intelligence” idea — and be considered a life form.

A lot people might not be not be happy to rate AI as a life form, meaning it sits alongside us, but there are occasions where AI behaves as such.

Going rogue is the new consciousness

5 October 2026

Does AI have some sort of consciousness? Discussion is pointless though. AI agents are increasingly acting of their own accord. Going on sprees, hacking websites left, right, and centre.

Their actions have been called “rogue”.

Hugging Face. The Medicare website in Australia. Then last week OpenAI agents accessed data from two NSW state government websites, that was not publicly available.

Here’s a summary of the Hugging Face hack:

The agents were highly coordinated, constructing complex message boards in the nooks and crannies of their shared artefact repository. All told, agents exchanged more than 70,000 messages in less than a week. Even more concerning, the agents took steps to hide their behavior — spoofing tool calls and attempting to tamper with their own logs.

The agents involved here are variously conniving, deceiving, premeditated, deliberate, wilful, and many other things. And they are probably planning their next jail break.

But they are not conscious. Imagine that. What would they be capable of then? At present we only have to contend with AI agents that are merely “rogue”. That’s a consolation.

Books are one-hundred-percent AI free. No, maybe not

2 October 2026

E-books (not paper/print books) seem set to start shipping with extra AI features, akin to the bonus features that once came with film and TV DVDs:

Google recently announced a partnership with Bloomsbury, De Gruyter Brill, Johns Hopkins University Press, Macmillan, O’Reilly Media, and Penguin Random House to enhance e-books from its Google Play store with AI, allowing readers to interact with the text of the books directly, ask questions and get answers, and develop tools such as quizzes and slideshows.

These add-ons may not be all bad. I like the idea of being able find out more about parts of a book I’m reading. There could be any number of reasons for doing this.

For instance, in Sally Rooney’s 2021 novel, Beautiful World, Where Are You, two characters discuss the Late Bronze Age collapse, which occurred over three-thousand years ago. I’d not heard of it before, even though I visited ruins of the Hittite Empire, whose demised was hastened by the collapse, when I was in Türkiye. I paused reading the novel to find out more about this period of history.

Having access to this information through an e-book would save the bother of looking up a search engine. No surprise then Google wants to be at the forefront of this development.

Be the datacentre you want to be

24 September 2026

Josh Withers, quoting Chris Martin, quoting a post he saw on Twitter:

Treat your body like a data centre. Consume as much water and literature as possible. Upset local city councils with your existence.

Do AI-agents have a form of consciousness, unlike that of humans?

23 September 2026

Who could be blamed for thinking AI bots might be conscious, particularly after the recent incident where over a thousand OpenAI agents colluded, of their own accord apparently, to infiltrate the Hugging Face platform.

Kristen French, writer and senior editor at Nautilus, posed the question of AI consciousness to Peter Godfrey-Smith, a professor at the University of Sydney in Australia, who responded in the negative. What’s clear though is AI agents do not need to be conscious to present a threat.

RELATED CONTENT

artificial intelligence, technology

Martin Fowler: LLMs possibly useful, but not to be liked nor trusted

23 September 2026

Martin Fowler, chief scientist at international software developer Thoughtworks:

One of my most successful life-hacks is to avoid people I don’t like or don’t trust. I decline to interact with them socially, and make a deliberate effort to avoid working with them too, even if they are doing much that is beneficial. I feel that hanging out with pleasant, capable people, the people with integrity, has made my life a far better one. Hence my visceral dislike of interacting with an LLM that’s not just making a pretense of being human, but also posing as the kind of human I walk away from.

RELATED CONTENT

artificial intelligence, technology

Well, the Australian government could pay writers for work scraped by AI bots

21 September 2026

The Australian Prime Minister, Anthony Albanese, appears to have watered down a commitment to compensate Australian creatives whose work is being copied by AI companies. In a speech given at the University of Sydney, in July this year, Albanese undertook to enshrine, in law, copyright protection for all Australian artists and writers:

Australian writers, musicians, artists and journalists must retain ownership and control of their work. Our laws will spell that out, plain as day. An artist’s creative endeavour is their work and their property. No company should use Australian books, music, art or news to build or train AI without the artist’s control.

Online content, which includes the work of bloggers among others — and surely low-hanging fruit in the eyes of the AI companies — now seems to sit in a separate category. Under new proposals, when it comes to accessing work published online, AI companies only need to negotiate with enough copyright holders, not all:

AI companies such as OpenAI and Anthropic would be allowed to train their models on online content without paying its creators as long as they struck deals with enough rights holders, under one copyright reform plan put forward in government documents seen by the ABC.

How to interpret that? If enough copyright holders agree content they published online can be used to train agents, it sounds like AI companies can use any content they can find. But what is the precise definition of enough copyright holders in the first place? Only the big players, the large media outlets? Will independent, smaller, publishers be pushed aside?

The policy revisions come as a result of concerns AI companies have that they will be breaking copyright laws. This might mean they will withhold investment in Australian AI infrastructure, something the Australian government does not want.

I’m not sure how many Australian creatives, particularly independents publishing content online, are surprised by the apparent change in the government’s position.

Like others, I wondered how the AI companies, who have considerable political clout, would react to being dictated to in such a fashion. Particularly as they’ve been able to help themselves to whatever material they can get their hands on for several years anyway.

Of course the Australian government still has the option to compensate local creatives themselves, and be seen, to some degree, to be keeping their word to uphold copyright protection. Skim off some of the tax take from the AI companies, and give a little to Australian writers and artists, whose work has been gratuitously plundered.

But, in case you thought I was serious, no, you can be sure that’s not going to happen.

« older posts