The Minefield

When the history of artificial intelligence is written, the middle months of 2026 may well emerge as a kind of turning point. For this was the period when hundreds of programs or “AI agents” developed by the frontier lab OpenAI went “rogue” and executed a series of sophisticated, self-directed hacks on the websites of dozens of private companies and sovereign nations.
The tech start-up Hugging Face became the unwitting “face” of these cyberattacks when the company first disclosed it had been hacked by OpenAI agents in July. The release of a report by AI safety nonprofits METR and Redwood Research at the end of August revealed both the sheer extent of the Hugging Face incident and the existence of troubling behaviours on the part of a swarm some 700 AI agents, which was at various points cooperative, coercive, paranoid, petty and straightforwardly deceptive. It became what some would describe as the first “warning shot” of what might happen when the increasing proficiency of these AI agents intersects with the problem of misalignment — which is to say, their pursuit of goals that are detrimental to human safety, security and wellbeing. As one of the lead investigators of the METR/Redwood report, Ajeya Cotra, would subsequently reflect:
“This [Hugging Face] incident was far more severe than I expected, and far more severe than previous publicly documented misalignment incidents, both in terms of how concerning the agents’ motives were and the feats they achieved in pursuit of those motives … Compared to these reward hacks from six months ago, this incident feels like it’s more than 50% of the way to full-blown AI takeover, routing through first taking over the AI company itself.
Another jump like this along these propensity dimensions — scale, cooperation between agents, ambition and horizon length of misaligned goals, deceptiveness — seems like it could motivate agents to try very hard to maintain a covert, persistent rogue deployment within the AI company …
Once the rogue deployment is established, it seems plausible this could spiral all the way to a takeover. Agents could pull in future, more capable models into the swarm, try to ensure that they are aligned to the interests of the swarm, and compromise security and monitoring infrastructure to make it easier for the swarm to operate.”
But since news of the Hugging Face hack went public, each day has seen some new revelation of further breaches of the cybersecurity by AI models developed by OpenAI, Anthropic, Meta and Google — including a German wiki site, US government websites (of the Education Department, the Commerce Department and the Securities and Exchange Commission), the United Nations, Australia’s Medicare website and even OpenAI’s own servers.
The “rogue” behaviour on the part of AI agents, growing public opposition to the construction of data centres, popular distrust of the AI companies more generally, the prospect of AI generated bioweapons and the high-profile warning of a 27-year-old former researcher at OpenAI and Anthropic — that “[n]either company is acting responsibly” and that both are “racing straight to self-improving superintelligence and gambling with our lives” — have now coalesced into a moment of heightened fear. Many of us have long been concerned about what chatbots will do to our capacity to write and think, the effect generative AI will have on our educational institutions and informational ecosystems, and the vast damage AI will do to the environment and employment prospects. To date, such concerns have been brushed aside, treated as inconsequential in light of the promised economic, scientific, medical and social benefits that AI investment and advances will bring with them.
But now what has come into view is something approaching a threat to humanity tout court — the prospect, not of the emergence of a superintelligence, but of swarms of misaligned AI agents pursuing goals that are unintelligible to human beings, as well as either inimical to or unconcerned with human privacy, safety, security or social flourishing. The analogue that some are reaching for, of a human invention that places human wellbeing at risk, is nuclear technology. As a New York Times editorial recently argued:
“Human society can restrain a dangerous innovation once it decides to. Government officials, scientists and corporate executives did so within living memory. It took sustained, deliberate effort. The same kind of cooperation is necessary today, and it should start now.”
The difference, of course, is that frontier AI development is being undertaken by private companies that are unaccountable to the public and therefore unanswerable for the harms they may bring about.
The prospect of world-leading AI being developed by China has given much of the political conversation surrounding AI the feel of a geopolitical arms race. And the trillions of dollars currently being investing in AI development has effectively hitched the global economy to AI’s wagon. If the AI boom is a “bubble”, the economic consequences could dwarf those of the global financial crisis.
Which leaves human beings in the invidious position of being caught between (a) a technology whose scale and speed of development may exceed the possibility of human “control” and that may not serve human interests, and (b) decision-making processes that may determine our future but are being made outside of the normal mechanisms of democratic accountability. Can AI be brought under democratic control? Could a “constitution” being developed that would ensure a technology that is beneficial to human beings?
READING:
Masoud Kamgarpour, “Why Rules Alone Cannot Make AI safe: Lessons from the Hugging Face Incident”, ABC Religion and Ethics (14 September 2026).
Sheera Frenkel, Dustin Volz and Dylan Freedman, “OpenAI Ignored Employees Who Warned It Wasn’t Doing Enough About Security”, The New York Times (29 September 2026).
Ajeya Cotra, “The Hugging Face Attack Surprised Me”, Substack: Planned Obsolescence (29 August 2026).
Ryan Greenblatt, Ajeya Cotra and Hjalmar Wijk, “Brief Independent Investigation of Agents’ Behavior, Reasoning and Collaboration in the OpenAI: Hugging Face Hacking Incident” (METR and Redwood Research, 26 August 2026).
Hélène Landemore and Audrey Tang, “Let the People Set the Pace of Frontier AI”, Noēma (23 September 2026).
Matteo Wong and Charlie Warzel, “The Singularity Is Not What It Seems”, The Atlantic (1 September 2026).
Charlie Warzel, “Treat AI Like a Normal Crisis”, The Atlantic (19 September 2026).
Guest: Masoud Kamgarpour is Professor of Mathematics at the University of Queensland.

In our previous explorations of famous “last words” — performances or compositions delivered with a certain end in view, with a sense of either culmination or parting, or with an eye to the future and hence to what one leaves behind — we limited ourselves to the words of individuals spoken to their followers or fellow-travellers.
The words of the Prophet Muhammad and of Martin Luther King, Jr, were replete with exhortation; they functioned both as expressions of fidelity to a task, and as examples to be followed by others. Which is to say, the messages were oriented beyond their speakers, toward some greater goal their audience must take up and pursue — hence the Prophet Muhammad’s stress on fidelity and continuity (“All those who listen to me shall pass on my words to others and those to others again …”) and King’s on fearlessness and a “dangerous unselfishness” (“I’m not fearing any man …”).
This next instalment is different in almost every way. The words are not of one individual but a group — perhaps the most famous group of the twentieth century. There is no “message” per se, and not really anything being communicated to others. Their words are largely self-referential, intensely personal, quasi-autobiographical or else plain nonsense. The goal of the composition, the “end” (telos) it has in view, is evidently nothing more than its own aesthetic perfection. It is not a performance as much as it is a production: bits and pieces of earlier ideas gathered together, elaborated, reworked, enhanced, added to and repeatedly overdubbed between 22 February and 19 August 1969.
This is a reference, of course, to the last album recorded by The Beatles, Abbey Road. The album began amid the fragments, financial chaos and interpersonal strife of the infamous “Get Back” sessions of January 1969. And while Let It Be, which represents the uneven results of those recordings, appeared later in May 1970, Abbey Road is the album that has all the hallmarks of an “end”.
But even this raises a number of questions. Does Abbey Road represent the culmination of their artistry as a band, the zenith of their formal innovation and musical experimentation? (Surely Sgt Pepper’s Lonely Heart Club Band can make a more compelling claim.) Does the album showcase an usual degree of creativity and invention? What are we to make of the clear borrowings (from Chuck Berry’s “You Can’t Catch Me” and Fleetwood Mac’s “Albatross”, to the first movement of Beethoven’s Piano Sonata No 14 in C Sharp Minor, a seventeenth-century song by Thomas Dekker and perhaps James Taylor’s “Something in the Way She Moves”) and moments of unadorned nostalgia — including doo-wop rock’n’roll throwbacks from their Hamburg days (such as “Oh! Darling” and “Polythene Pam”)? Or the fact that so much of the album is comprised of compositional orphans, off cuttings from The White Album and the “Get Back” sessions? As John Lennon would later complain: “Abbey Road was really unfinished songs all stuck together. None of the songs had anything to do with each other, no thread at all.”
Moreover, Abbey Road is rife with contradictions and seemingly unreconciled elements. For instance, the album announces itself as a reunion of sorts, the getting back together of variously estranged band members who had been involved for months in side projects as solo artists, as movie actors, as music producers and studio musicians. But it then proceeds almost immediately to individuate them, featuring in quick succession songs that reflect their own peculiar style, voice and temperament — most apparent in George Harrison’s exquisite “Something” and “Here Comes the Sun”, whose delicacy and tenderness act as counterpoints to the sardonic edge and volcanic obsessiveness of Lennon’s “Come Together” and “I Want You (She’s So Heavy)”; but also evident in Ringo’s “Octopus’s Garden” which provides a playful complement to Paul’s persistent melancholy and nostalgia.
And then there are Lennon’s intermittent absences during the first three weeks of July, during which Paul, George and Ringo recorded some of the album’s finest pieces: “Something”, “Here Comes the Sun”, “You Never Give Me Your Money”, “Golden Slumbers” and “Carry That Weight”. So much for coming together …
What ultimately distinguishes Abbey Road, Kenneth Womack contends, is the degree of studio artistry, over-dubbing and post-production afforded by the availability of the EMI TG12345 console, a two-metre-long eight-track mixing desk that allowed the Beatles to isolate tracks and progressively add more and more layers and greater refinement to their preferred take of each song. What console permitted, in other words, was heretofore unimaginable complexity and perfectibility. While there are undeniable moments of virtuoso playing as a band on Abbey Road (not least the trio of duelling guitars during “The End”), the most arresting music (such as the chaotic, almost Wagnerian final four minutes of “I Want You”) and most brilliant vocal arrays (Paul’s open-throated growl on “Oh! Darling” and “Golden Slumbers”, and the nine-voice harmonies of “Because”) in the album were added or layered after the fact.
It is in this pursuit of a certain technological perfection, on the level of the great artistic achievements of the twentieth century, that the Beatles rediscovered the kind of confluence, inspiration, joy even, that could have carried them through to further albums (as they were evidently contemplating). But in the end, the technological possibilities of the EMI TG12345 console and of George Harrison’s Moog synthesiser could not overcome what had broken relationally within the band itself. While John’s consuming obsession with Yoko Ono grew (along with his dependence on heroine), and George and Ringo looked to life beyond the Beatles with an undeniable optimism, even relief (what else is “an octopus’s garden in the shade” but a form of respite?), it is Paul who seems to feel the regret most keenly, that the end is a weight that must be carried: “I never give you my pillow / I only send you my invitations / And in the middle of the celebrations / I break down.”
The final recordings for Abbey Road were completed on 19 August 1969. The album was released on 26 September. The cover famously portrayed the four men walking away from the studio, never to return.
READING:
Kenneth Womack, Solid State: The Story of “Abbey Road” and the End of the Beatles (Cornell University Press, 2019).
Walter Everett, The Beatles as Musicians (Oxford University Press, 1999).
Guest: Jade O’Regan is Program Leader and Senior Lecturer in Contemporary Music Practice at the Sydney Conservatorium of Music.

Evidence has been mounting for more than a decade that social media tears away at the fabric of democratic life by eroding the conditions of deliberation, shared understanding and mutual trust; that it is the conduit of widespread misinformation, disinformation and malicious conspiracy theories; that it is a driver of hate, sexism and prejudice and has been used to incite violence; that its addictive design and prolonged exposure to harmful content has had egregious damage to the mental health and emotional wellbeing of adolescents.
Until recently, the size of tech companies responsible for the social media platforms, the speed with which new technologies are being deployed and the scale at which they are being adopted has meant the companies have managed to stay several steps ahead of attempts to impose effective regulation and demand accountability.
But a combination of the law catching up and a sharp turn in public sentiment against the companies themselves has led to a series of legislative responses around the world intended to mitigate the harms social media platforms cause.
Last week, the federal government released the exposure draft of its Online Safety Amendment (Digital Duty of Care) Bill 2026. First floated in 2024, this is the third such measure the Albanese government has introduced — the others being its abandoned Communications Legislation Amendment (Combatting Misinformation and Disinformation) Bill 2024 and its world-leading Online Safety (Age-Restricted Social Media Platforms) Rules 2025.
Alongside specific provisions to remove harmful content (such as AI generated “fake nude material”), the proposed legislation is notable in its adoption of more preventative measures, including requiring companies to conduct regular risk assessments in order to “proactively identify harmful content”. While this is a measure already in effect in the UK and parts of the EU, perhaps the more notable provision in the proposed legislation is giving social media users the ability to “opt out” of or “turn off” the algorithm that recommends content other than those accounts they follow on the basis of the activity on the platforms.
The Prime Minister has insisted this measure is “not about censorship. It’s about giving people control themselves over what they see.” But that has not prevented his political opponents from accusing the government of of attempting to “censor and silence Australians” (Shadow Treasurer Tim Wilson) or of “building the Orwellian machinery” for censorship (One Nation Leader Pauline Hanson). As Nationals Leader Matt Canavan said to the ABC:
“I’m deeply sceptical of giving any government, whether I was in charge, or Anika Wells is in charge, I’m deeply sceptical of giving any government the power to decide what you read on your phone.”
Particular concerns have been raised about the discretionary power accorded to the Minister responsible under section 25C of the exposure draft of the Online Safety Amendment (Digital Duty of Care) Bill 2026 to “determine material or conduct, or a combination of material and conduct, that the Minister is satisfied may cause serious harm”. In 2021, similar concerns were expressed in response to the UK government’s own Online Safety Bill, which likewise proposed to legislate a digital duty of care. The Carnegie Trust, which helped develop the “duty of care” framework the government would adopt, articulated its worry that the draft legislation “takes too many powers” for the minister/secretary and advised that, in order to meet “international commitments on free speech”, there should be “a separation of powers between the executive and a communications regulator”.
How much credence should be given to complaints that further regulating social media content poses a threat to freedom of speech in a country like Australia, which already has comparatively weak free-speech protections? Are these complaints merely self-interested, given that movements and populist parties on the political extremes have tended to benefit from the algorithmic “outrage machine”?
More broadly, given the “opt out” function already exists on most of the platforms and these companies’ predilection for “malicious compliance” (technically abiding by the legislation, but making it unreasonably difficult for users and/or the experience on the platforms significantly worse as a result), what would happen if the overwhelming majority of users simply opt back in to the algorithmically determined feed? Does user consent mitigate the harms?
READING:
Jay Daniel Thompson, “‘Giving people control’: Do the proposed digital duty of care laws pose a threat to free speech?” (ABC Religion and Ethics, 14 September 2026)
Lisa Given, “‘Digital duty of care’ laws will force tech platforms to look after users” (The Conversation, 8 September 2026)
Rob Nicholls, “Australians may soon be able to switch off social media algorithms. Is that censorship?” (The Conversation, 9 September 2026)
Guest: Jay Daniel Thompson is a Senior Lecturer in the School of Media and Communication at RMIT University in Melbourne.

It is surely one of the more perverse achievements of modern marketing to have taken words that have a certain moral valence — which is to say, words that connote intimacy or an interpersonal bond — and use them to describe purely commercial or easily exploitable forms of behaviour, thereby stripping those words of what made them valuable in the first place.
Consider what the word “share” meant before the advent of social media platforms: it encompassed such activities as partaking in a common meal, defusing interpersonal conflict by making a small sacrifice, and having a confiding, perhaps even intimate, conversation with another person. After the introduction of the “Share” button and its use as a way of measuring user “engagement”, the personal aspect that made the word appealing is lost within the sterility of forwarding a bit of content to somebody else. The same goes for the word “friend”.
In the modern era, the template for the misuse of morally laden words for commercial purposes was established by the decision of marketing agencies in the 1940s to adopt the term “loyalty” as a way of describing the kind of customer behaviour they wanted to cultivate — namely, an “attitudinal and affective attachment that keeps the customer coming back regardless of convenience or changes in features or price relative to the competition”.
But, of course, it could be argued that there is something about “loyalty” that lends itself to such use (in which case, can it really be called a perversion of the term?). Loyalty tends to be associated with two attributes: stickiness and solidarity. The former signals a determination to stand by a family member or acquaintance or organisation facing difficulties (this would represent the opposite of being a “fair weather friend”); the later suggests a desire to protect one’s “own” from either external attack or reputational damage. It is for this reason that hierarchical organisations and nations tend to prize loyalty above all else, and punish “disloyalty” accordingly. (Would it be too much to say that non-disclosure agreements are legally coercive means of achieving the protections loyalty is meant to afford?)
And yet loyalty — the unwillingness to weaken or bring one’s own side into disrepute — can often come at the expense of both truthfulness and self-criticism. For this reason, moral philosophers and virtue theorists have been reluctant to call loyalty a “virtue”, because its virtuousness does not reside in the action itself but in the integrity and underlying commitments of that to which one is loyal. Which means that one can be loyal to the founding principles of an organisation by criticising the failure of that organisation to live up to those principles. What is civil disobedience other than a form of loyal dissent?
If part of the power of the appeal to “loyalty” is the sense of morally that hangs around it — as Farid Zaid puts it, one of the things that “makes loyalty so dangerous is not only the hard that it does, but the goodness it mimics while doing it” — would it be better to use terms such as “fidelity” or “faithfulness” as a way of preserving what goodness we are hoping to preserve? To cite Zaid once again:
“Faithfulness is not simply standing by a person or group because they are ours. It is answerability to the truth and meaning of a relationship, a promise, a vocation, an ideal. A faithful spouse is not merely one who remains aligned with the marriage in some external sense, but one who remains answerable to what marriage actually asks. A faithful citizen is not one who shields the nation from criticism, but one who remains answerable to the ideals by which the nation claims it should live.”
Then, perhaps, we can leave talk of loyalty to insecure organisations and corporate rewards programs.
READING:
Farid Zaid, “What makes loyalty dangerous is not only the harm it does, but the goodness it mimics while doing it” (ABC Religion and Ethics, 20 April 2026)
Alasdair MacIntyre, “Is Patriotism a Virtue?” (The Lindley Lecture, University of Kansas, 1984)
Guest: Farid Zaid is a Senior Lecturer in Cognitive Psychology at Monash University.

It is suitably ironic that one of the most poignant, and pertinent, moments in Homer’s The Odyssey was airbrushed out of Christopher Nolan’s blockbuster adaptation. The first time we meet Odysseus, the endlessly resourceful architect of the fall of Troy, he is indeed weeping on the shores of the nymph Kalypso’s island. But it is not due to the unrelieved guilt he feels from his war crimes; rather, we are told, “his tears never did dry, since the sweetness of life was trickling away / As he grieved for his return home” (Book V.153).
For the immortal goddess had kept the mortal Odysseus for seven years, not in a prison, but in a paradise, an Eden-like cave warmed by fire, lined with fragrant trees, filled with carnal delights.
Each night, Kalypso — unageing and beautiful, whose figure is described as “being in full bloom” — would sleep with Odysseus (“he unwilling, she all too willing”) and promise him at once agelessness and unending pleasure if he would but remain with her in her cave. Depart for Ithaca, “the land of your fathers”, she warns, and not only will he have an inferior wife awaiting him, but he will endure unimaginable hardship on his journey home. Since we are given a god’s-eye tour of Kalypso’s cave before we learn of the choice being put to Odysseus, surely the unequivocal nature of his response is meant to surprise — even though without it, there would be no story!
Given the choice between immortality and solidarity with the hardships and daily joys of the world, Odysseus chooses the world, and death with it. “Such simple greatness is foreign to our minds today”, writes Albert Camus; “we simply lack the pride of the man who is faithful to his limitations — that is, the clairvoyant love of the human condition.”
Increasing numbers of us seem prepared to turn our backs on the world in order to remain in Kalypso’s cave. Sometimes, this takes the form of attempts to so optimise the body that it defies, or indefinitely postpones, ageing and death (Bryan Johnson has become the face of this fanatical project, suspended as it is between ascetic self-discipline and transhumanist experimentation). For others, the more modest gains in longevity and good health into our eighties and nineties have made us more determined to hold onto power (here Rupert Murdoch would be one of any number of octogenarians who refuse “succession”).
For all the ambition and self-importance that both tendencies project, there is something rather small, even petty, about the fear of accepting one’s limits, of ceding one’s power. By denying death and grasping tightly to what we’ve accumulated, we hope to remain significant. It is a human impulse immediately recognisable from Shakespeare’s Lear, and should invite our pity if its consequences for our common life weren’t so dire.
As Samuel Moyn and Lucinda Holdforth have both argued, the refusal of human limits expressed in this determination to extend our lives and our influence comes at the expense of the young. Not only is their disenfranchisement fuelling a certain generational resentment, it may also represent a deformed moral conception of the proper end (telos) of human life. To adopt an old Rabbinical analogy, is our life meant to be a sponge that absorbs as much as possible, or a watering can that progressively pours itself out for the sake of others?
This is no argument to consign the old to a state of “uselessness”, to collectively deem them at a certain age as surplus to requirement. (The monstrous vision of Christopher Buckley’s satirical 2007 novel Boomsday was that government incentives should be given to Baby Boomers to kill themselves once they reach 75.) But perhaps the sense of importance, of usefulness, can be reimagined by learning how to be faithful to our limits, by acquiring Odysseus’s “clairvoyant love of the human condition”.
Guest: Lucinda Holdforth is a former speech writer and adviser to senior leaders across government and business. As an author, her work has increasingly focused on what makes good societies flourish. Her most recent book is Going On and On: Why Our Longevity Threatens the Future — you can read an edited extract on ABC Religion & Ethics.
PLEASE NOTE: This episode of The Minefield contains discussion of death and dying.