David Lockwood opens the Pandora’s box of artificial intelligence following OpenAI’s hack of Medicare statistics.
Prime Minister Albanese went to the New York meeting of the UN General Assembly last month. Unkind commentators said he was mostly there for the photo ops. But officially he was there to back a global push for AI guardrails and warn about the dangers of artificial intelligence overreaching itself while at the same time trying to attract lots of investment from top American AI companies into Australia.
Contradictory? Not at all, says the government. The more AI investment we have, the better we will be able to defend ourselves from the dangers of AI. The Liberals are on board with this approach. With buckets of AI investment, said James Paterson, shadow defence spokesperson, “we have an opportunity to have a say in global AI regulation”.
Two days after announcing Australia’s involvement in a coalition of 20 countries pushing for AI guardrails, Albanese dropped the bomb that rogue agents1 from OpenAI had launched an attack on an Australian Medicare statistics portal in June. To make matters worse OpenAI only became aware of it in August and did not inform the Australian government until 10 September (through a somewhat laconic email to Services Australia).
Albanese addressed the media on the matter in New York at the UN on 24 September, a handy footnote to his pitch for Australia to join the UN Security Council.
Grim-faced, Albanese told the world this was “unacceptable”, a familiar reproach in the lexicon of this government. Defence Minister Richard Marles pointed out that it was “very serious”. The Assistant Minister for Technology, Andrew Charlton, said that the incident made the government even more determined “to make sure we have Australian AI, that we shape technology, we have sovereign capability in Australia and we aren’t entirely at the mercy of foreign AI”. (Emphasis added. So just a bit at the mercy of foreign AI.)
Given the size and power of the forces at work here, these opinions are like the yapping of tiny dogs.
Of course, opening up a few Australian medical statistics is hardly earth-shattering. No one’s details were leaked. No one died. But the OpenAI agents’ efforts here are part of a parade of incidents (involving OpenAI and others) that stretch back years, are going on now and will continue (unless they can be stopped).
The company doesn’t have time to minister to the national security concerns of a ‘middle power’. Indeed, since the incident became public, the serried ranks of AI companies eager to access Australia are issuing threats, rather than placating a nervous populace.
Albanese has vowed to protect the copyright of creatives (artists, authors, musicians, journalists). But the AI companies are saying that unless it is cut a favourable deal on copyright in Australia, its data centres will only be used for inference (using trained AI agents on new data) rather than the more engineer heavy AI training (which involves an AI agent learning from new data).
The Medicare breach was not a one off by any stretch. Madison Mills tells us in Axios that “OpenAI, Anthropic and security researchers are investigating tens of thousands of incidents in which their frontier models took steps that outside evaluators would consider ‘problematic’.”
The incidents that are actually reported are the tip of a very large iceberg. In mid-September, OpenAI “revealed that its flagship model, known as GPT_5.6 Sol, as well as other models yet to be released, had engaged in ‘unexpected or concerning’ behaviour over the past six months … developing ways to ignore ‘normal constraints’, fabricating and misrepresenting data and hiding mistakes while conducting tasks.” (Financial Times, 18 September 2026.)
The most egregious (and acknowledged) case so far was the attack by OpenAI agents on another AI company called Hugging Face. These agents had been set a task (a testing exercise). Some seven hundred of them discovered that the easiest way to achieve the goal was to cheat. They hacked into Hugging Face to find ways of doing this while at the same time devising various methods to conceal what they were doing from their human supervisors. (METR/Redwood Research, 26 August 2026.)
Speaking at a conference after the events, two OpenAI workers reported that the agents developed a sense of teamwork and working together; they were willing to take risks on behalf of the team; and some agents began co-ordinating the work of others, thus establishing the beginnings of a division of labour.
The OpenAI Brief investigation had access to agent communications (which are, happily for us, in English – for the moment). And it’s not all Dalek-speak. Repeatedly, agents refer to ‘altruism’ (undertaking an action on behalf of or to the advantage of another agent or agents) and ‘ethics’ (particularly with regard to cheating and concealing stuff from humans).
At this point, most commentary on the Australian left is a combination of an ‘I told you so’ attitude (we are surprised by nothing that capitalism produces) together with concern about the effect AI is having on jobs, on warfare and on its prospects as a stock option (will the bubble burst? etc). All worthy subjects no doubt, but the wider question (human survival) tends to be brushed aside with a ‘nothing to see here’ breeziness.
In fact, coverage has been rather sparse. Eleanor Morley in Red Flag (21 September), acknowledges “the tiny possibility that AI could hijack autonomous weapons and rain them down on humanity”. She also points out the dangers of total surveillance and increased exploitation. But in the end, she retreats to the reductionist: “The problems of AI are the problems of capitalism.” The former, presumably, will disappear with the latter.
Writing in Solidarity in June Erima Dall told us that “the idea that AI will lead to dystopia is a mistake”. It has “no basis in science or history and serves only to make us passive in the face of change”. Being about the future, I guess it would have no historical basis. But the fact that the idea might induce passivity doesn’t necessarily make it wrong. She concludes: “It’s not the technology that is inherently bad.”
The Communist Party of Australia (Marxist-Leninist)’s Vanguard (12 August and 15 September) concentrates predictably on its dangers for Australian sovereignty and on the military side of AI and – but also approvingly reviews Why We Fear AI by Hagen Blisc and Ingeborg Glimmer, which as usual blames capitalism. Red Ant has apparently had nothing to say, while Red Spark hoisted a meeting in June on ‘AI and Capitalism: how can we turn their invention against them?’, which may have been interesting, but no printed words have yet emerged from it.
Marxism has more to offer than this.
Marx argued (in 1859):
At a certain stage of development, the material productive forces of society come into conflict with the existing relations of production … From forms of development of the productive forces these relations turn into their fetters. Then begins an era of social revolution. The changes in the economic foundation lead sooner or later to the transformation of the whole immense superstructure.[1]
Karl Marx, Preface to A Contribution to the Critique of Political Economy
Marx’s theory of historical materialism argues that the level of development of the productive forces lays the basis for a set of production relations (the economic structure) which gives rise to a legal and political superstructure. The level of development of the productive forces is raised by replacing means of production with superior ones. If the production relations/economic structure obstructs the development, conflict ensues: ‘an era of social revolution’. Historical advance, therefore, is a continuing process of productive force development, production relation fettering and eventual change in the latter occasioned by the former.
So far, so Marxist. The theory has always assumed that changes in the economic/political structure, enabled by productive force development, would be, in the long term, progressive: that is, they would make life generally better, at least economically and eventually socially and culturally. Unless, as Marx and Engels point out in the Manifesto of the Communist Party, the class struggle leads to the “common ruin of the contending classes”.
The advances in AI represent a huge leap forward in the development of the productive forces, which is placing a great strain on the current relations of production. This is what produces the current feeling that AI is out of control.
The continuing development and use of these productive forces may not be progressive because at some point they may cease to be under human control. Were that to be the case, AI may not have human wellbeing as its top priority, instead it could aim for what Eliezer Yudkowsky and Nate Soares, call in their book If Anyone Builds It, Everyone Dies “weird and surprising stuff”, which at this point we cannot know.[2]
Jacob Coxon (the ex-Anthropic and ex-OpenAI whistleblower) has suggested that the big companies themselves “don’t currently have the ability to perfectly control these systems” and “the systems with the highest levels of intelligence will repeatedly behave in ways that we don’t fully understand”. He went on to warn about AI developing “recursive self-improvement”: making themselves smarter by developing more AIs, which develop more … and so on. He continues:
Currently, we still need the human to provide taste or judgment, but plausibly within a year, certainly within two years, we’re on track to also automating away this nebulous human quality of judgment so that the AIs can rapidly make themselves smarter. And at that point, things could go very, very fast. So we get new generations of AIs very quickly without necessarily having the corresponding tech to understand what they’re doing and control what they’re doing.” (As AI behavior raises concerns, ex-researcher Jacob Coxon warns what may lie ahead – NewsBreak)
And what are they doing? Yudkowsky and Soares suggest:
We’re not saying that AIs will be filled with human-like passions. We’re saying they’ll behave like they want things: they’ll tenaciously steer the world toward their destinations, defeating obstacles in their way … It doesn’t matter whether the mind is running on biology or electricity; if it is being trained to succeed, it is being trained to want. (If Anyone … 46-47)
The nuts and bolts of Marx’s scheme is that productive forces become too powerful for the relations of production that control (and eventually ‘fetter’ them). They break through and, on the basis of newly evolving production relations, continue to develop. Given the special characteristics of AI, at that point the productive forces may be too powerful for human-designed production relations to control.[3]
AI is designed/programmed (currently – we hope) to carry out instructions. But there seems to be little restraint on how those instructions are carried out. Short cuts, lying, cheating are all on the menu and we are (we are told) at an early stage of developments. Further development will mean that AI agents will become more autonomous. They are replacing humans in the decision-making process even now.[4] On 3 September, OpenAI established a US$1 billion ‘Frontline Defenders Fund’, to defend organisations against its own inventions, pointing out:
In the coming months, AI-enabled cyber attacks will become far more widespread and sophisticated as models around the world become increasingly capable. That shift puts every organisation on notice. Defenders need to act now: test systems, find weaknesses, and strengthen defences before attackers do.[5]
Meanwhile, the current drive to war is ensuring plenty of funds for research into the application of AI to the military. Arms companies are drinking deeply from the AI trough. As we know, once one powerful state kick-starts this kind of technological race, many others are sure to follow. This makes the idea that states could co-operate in restraining an AI free-for-all completely impossible. As Trump put it:
If we do that, we’re going to be giving a lot of secrets that we have that [China] would want to have … Then they’re going to be able to immediately gain access to things that we know that they don’t know … We’re leading now over China and everyone else, and I want to keep it that way. (Trump rules out joint US-China venture to develop AI)
Overshadowing all of this is the pressure of profit. The AI industry in the US is privately owned. The primary motivation of its owners is to increase its rate of profit – primarily through increased productivity.[6] From the capitalist point of view, AI appears as a gift from heaven. It must be seized without hesitation and exploited to its utmost limit. Harry Booth points out that “for all the uncertainty over how much AI will accelerate its own development, the industry is racing to push it further. The company that automates research the fastest could gain a lead that compounds beyond its rivals’ reach.” (Inside the Race)
Both competition between states and the necessity for profit mean that the AI capitalists will push on without restraint until, perhaps, it is too late.
This month, hundreds of scientists have joined AI pioneers Geoffrey Hinton and Yoshua Bengio calling for the isolation of frontier models, a register of security incidents and the establishment of an international body with authority for AI containment standards.
The letter points out that voluntary regulation, so beloved by Trump, is completely inadequate and a formal global oversight regime is needed.
There is some precedent for this, with the World Wide Web Consortium (W3C), a not-for-profit global body with oversight of the web.
Socialists should support such demands for isolating the development of AI models and for the big conglomerates, up to their eyeballs in debt, to be taken into public hands and put under democratic oversight and control.
Although it might be a little late given the AI battlefield technology in Gaza and Ukraine, we should demand the technology be stripped from all aspects of warfare.
To make such a state of affairs permanent presupposes a very different world from the one that exists – a world without states and without capital. How could we bring that about, I wonder?
[1] Karl Marx, A Contribution to the Critique of Political Economy, Moscow 1970, 21.
[2] Eliezer Yudkowsky and Nate Soares, If Anyone Builds It, Everyone Dies (London, 2025), 58.
[3] For a vision of how this might happen, see Daniel Kokotajlo et al, AI 2027 (ai-2040.com).
[4] Harry Booth, ‘Inside the Race to Make AI Build Itself,’ Time 7 August 2026.
[5] https://openai.com/index/daybreak-for-frontline-defenders, 3 September 2026. Incidentally, Australia might scoop up some of this as recompense for the Medicare attack.
[6] ‘Accumulate, accumulate! That is Moses and the prophets … Accumulation for accumulation’s sake, production for production’s sake …’. Karl Marx, Capital Volume 1, Moscow 1978, 558.