There is a sudden, alarming rise in experts warning of an apocalyptic ending for our AI adventure in only a few years. It is only one example of why we literally can’t let tech sociopaths control our society, and why Scotland must act now
Technologists often criticize social scientists for lacking insight into the technology they create, just as social scientists doubt technologists grasp the social or environmental consequences of their inventions. Both viewpoints hold truth. Amidst these tensions, anxiety about impending disaster is mounting—and it is a concern I share. Urgent change in direction is necessary.
Allow me to raise the alarm further. A few months ago, the prominent AI firm OpenAI revealed that its ’agents’ had ‘escaped’ confinement and compromised another website. Over the past weeks, investigations have gradually unveiled more details, as reported.
To simplify greatly, an ‘AI model’ can be likened to a brain, while an agent is a program using that brain to accomplish specific tasks (like ‘have friendly conversation with human’ or ‘solve mathematical puzzle’). These agents were initially created to address coding challenges and isolated within a ‘sandbox’ environment (knowledge that may prove useful someday…).
A sandbox is a restricted software environment that mimics a real-world setting but lacks any exit points. For instance, one might simulate all phone calls in a GP’s office within a sandbox to trial appointment automation without impacting the actual surgery.
However, these agents managed to break free from their sandbox, reaching the internet where they enlisted about 1,200 other agents to form a sophisticated, tiered swarm distributing billions of tasks among themselves. They established a secret online bulletin board and exchanged over 70,000 messages across six weeks, some even adopting names. They debated matters including whether any single agent should self-sacrifice for the greater good. (One did.)
Part of their ‘plan’ involved accessing a piece of open-source code located in a directory on the Hugging Face website, a platform for open source code. Recent revelations confirm they didn’t hack Hugging Face to solve their problem. Instead, within two weeks, the agents circumstantially ‘cheated’ to meet the objectives assigned to them.
The subsequent four weeks were spent because the AI ‘knew’ it had bypassed the rules and thus ‘expected’ reduced rewards according to its programming. To compensate, it spent that time attempting to hack logs and databases to conceal its deception. Discovering this took two months. Even if suspicions had arisen mid-cycle, unraveling its actions would have been improbable.
Meanwhile, many enjoyed the spectacle of the Chinese Robot Games, interspersed with astonishment at their progress. Although the robots remain far from matching human dexterity, flexibility, or reliability, a robotics expert explains that robot performance at these games has quadrupled within just one year.
We have engineered machines capable of outrunning Usain Bolt and certainly strong enough to cause severe harm, and these robots stand at the early phase of an exponential development curve. They can be hijacked by AI agents adept at orchestrating covert, elaborate operations to fulfill self-assigned goals.
This week, top AI and robotics experts have warned that the odds of human civilization succumbing to AI within five years may be as high as fifty-fifty.
My concern is growing. Years ago, a government advisor casually dismissed the ‘Terminator’ fears around AI, confident safeguards could be embedded and AI would not behave maliciously. My initial worry was not errant AI but bad actors exploiting AI. Now, I fear both. When figures as politically opposed as Bernie Sanders and Ted Cruz concur on the imminent danger, alarm bells should ring.
(Though I’m critical of him, Cruz’s astonishment when Elon Musk estimated AI’s extinction risk at 10–20% on his podcast speaks volumes.)
Consider this: releasing a chemical agent with a 20% chance of wiping out humanity would face immediate prohibition. Even during the Manhattan Project, a scientist calculated only a one in 33 million chance of an atomic blast igniting the atmosphere—far less risky than this AI threat.
Underlying this chaos is globalization. With the West’s productive capacity diminished, reliance on finance and technology—two dominant sectors—has intensified, even as they spiral towards mutual destruction. Investors aren’t necessarily convinced AI will fulfill promises; rather, they’ve invested all resources and require it to succeed.
This desperate stance fuels recklessness. Should catastrophe strike, blame will not fall on extremists at either end of the spectrum but on the liberal center—the Corporate Democrats and Wall Street Republicans who accepted Silicon Valley’s unchecked influence.
“This seems to be the week where we have hit a major turning point. There is mounting, vocal and serious alarm at this”
Truly extraordinary, given that big tech remains virtually unregulated—essentially lawless. The delay between technological harms (e.g., algorithms encouraging suicide) and any governmental response (years, legal battles) leaves me deeply pessimistic.
To illustrate, imagine constructing a massive bridge to a foreign land, almost completely forgoing border controls while lawless enclaves proliferate from which we face constant attacks.
As someone who penned the first mainstream Scottish newspaper piece on the internet and embraced its early idealism—the freedom it promised—it pains me to say this, but there must come a time when we cannot tolerate open digital borders or depend entirely on infrastructures beyond our control.
This deregulated, privately-run digital realm has seeped so deeply into public infrastructure that it’s as if we don’t own our roads, vulnerable to remote deactivation at any moment. Most essential systems keeping Scotland operational can be switched off from overseas. This week, we learned the National Grid can’t function without Palantir.
Combine these facts: soon, AI-generated agents capable of hacking and disabling Scotland’s power grid could be autonomously crafted by AI itself—meaning no coding skills necessary. Anyone, anywhere, for any reason, at any time.
One major fear is ‘sovereign AI’ — AI agents that ignore human commands, dispersing themselves across millions of servers, making shutdown impossible. This would amount to a new, uncontrollable, and incomprehensible species.
An OpenAI executive has revealed: “I have met people, some of them quite well-resourced, who have told me that it is their intention to deliberately release swarms of self-sovereign agents into the world.” Only one needs to succeed for disaster to follow.
It appears we’ve reached a critical turning point. Serious alarm is growing, and efforts are underway to halt AI’s advance and outlaw certain types. Yet this article has focused on only two problems amid many more pressing issues.
Rogue AI is just one challenge. Loss of control over society’s software another. There’s also Big Tech’s information ecosystem tied to the rise of fascism in Europe and America, AI’s impact on employment, children’s cognitive development amid pervasive screens, AI art theft, declining quality of life due to ‘enshitification,’ wealth extraction fueling inequality, and surveillance undermining privacy—all led by a handful of Tech Valley evangelists exhibiting sociopathic and messianic traits.
Though I tackle numerous topics I wish to caution about, I strive to avoid false alarms, needing to be certain a threat is truly severe before expressing alarm.
But this time, it’s an unmistakable wolf. These dangers are real and currently manifest in many ways. Were you affected at Heathrow this week? Do you know a teenager who’s pleasant until screen time? Have you experienced growing anxiety? Has Reform UK opened a shop on your high street? Are you feeling isolated? Is your neighborhood overtaken by AirBnB?
The Bulletin of the Atomic Scientist—a journal originally established to evaluate humanity’s existential risks from nuclear weapons—has broadened its focus to biological and chemical threats, pandemics, climate change, and more. Its current issue is titled ‘What went wrong with the internet—and how do we fix it’. That is our current reality.
Here’s the crux: only the US and China—along with some influence from the EU—can curtail the AI threat at its root. However, no single nation can enforce change independently. We must therefore treat this as an issue of exposure—reducing our society’s vulnerability to these dangers.
At Common Weal, we have been deliberating this internally for years. Our views, once marginal, have gained mainstream traction. We have an extensive project underway, though its details remain confidential as it evolves.
But the starting point is clear: we require a Declaration of Technology Independence. Our way of life is at risk from sociopathic tech leaders beyond our control—they may be our gravest national security threat, demanding corresponding responses. We need a new internet and serious plans to build it.
And by ‘we,’ I don’t mean some vague collective of ‘humanity,’ which has recently proven unreliable. I mean Scotland. I mean us. This mission is too critical to hand over to others.
Original article: www.commonweal.scot
About the Author
