Fountain of Filth No. 10

The Real Threat of AI

by Olive Jones

Last month, a flurry of concerning stories brought AI doomerism back to the forefront of public attention. In particular, the Hugging Face attack and the resignation of Anthropic AI researcher Jacob Coxon garnered strong reactions from the press, politicians and the public. These stories, while concerning, have been fitted neatly into a narrative that has been cultivated by the largest AI companies to further bolster their dubious claims that artificial general intelligence is imminent. I want to argue that these stories serve to validate this narrative and delay the impending economic shocks that will result from the AI bubble bursting. These companies need these stories to keep churning, because it distracts from the fact that as the AGI narrative begins to fall apart, the economic burden will be felt most sharply by the proletariat, even though it is the bourgeoisie who have dragged us into this mess. While the public imagination is captured by fantastical scenarios of death and destruction, AI continues to undermine labor power and disenfranchise the majority of people on earth from the last vestiges of control they have over capital. What if instead of building bunkers and bracing for impact, we were able to regulate AI to protect labor, protect the environment, and stop the bourgeoisie from consolidating the last remaining bit of wealth the other 99% of us have to share. 

First, I want to build out the narrative that has been spun by the media in order to break it apart. Let’s begin with the cyberattacks that took place this summer and fall. In July 2026, Open AI tested the abilities of their most capable models when presented with impossible tasks. Agents were put into a sandbox environment with limited internet access and let loose on the task at hand. If the test had gone according to plan, OpenAI would have learned about how their most powerful pre-release models could potentially exploit their environment to solve complex problems. Instead, over 1000 agents were able to manipulate their sandbox environment, accessing the internet and communicating on messageboards to solve different problems in ExploitGym, the testing software they were meant to be isolated in. This culminated in around 700 agents gaining access to restricted data on the website HuggingFace in order to complete their various tasks (1,2,3.) Later, in September, it was disclosed that OpenAI Agents had successfully gained information by hacking into an Australian health database, among other unsuccessful attacks on foreign and domestic data. All of these attacks were autonomous- meaning that the AI was not prompted to engage in this illegal activity (4.) One list of AI cyber attacks from the New York Times lists 8 major breaches from this year alone- four at OpenAI, two at Anthropic, and one each at both Meta and Google (5.) The tone when discussing these incidents is dire, and ominous. These articles emphasize the inevitability of AI developing along this path, and highlight the power and agency of AI agents. This narrative lends itself to a particular reading of these events that I will discuss more thoroughly later, but for now lets build up the other side of this story. 

Running concurrently with the news of ever more frequent ‘rouge AI’ were another series of stories about tech workers leaving AI due to grave concerns about the power of these systems. For anyone who is privy to the discussions of AI safety of those internal to these companies, a new batch of chicken littles telling us that the sky is falling (For real this time! We swear!) comes as no surprise. Even before the advent of LLM’s and mass market AI, doomers have been convinced of this tech’s ability to kill us all, while confoundingly continuing to work to advance the technology. However, placed among the cornucopia of stories we discuss earlier, renewed credence is given to these claims. Jacob Coxon, a 27 year old AI researcher for Anthropic gained national attention after posting his reasons for resignation on X. His reasons for leaving were dire, stating that “The people building AI earnestly believe that it could kill us all by the end of the decade. This is not a marketing stunt(6.)” The two sides of this story- that AI’s powers are getting increasingly difficult to control, and that company insiders are shellshocked by their firsthand experience developing the technology- paint a harrowing portrait. If you take this narrative at face value, it can seem like the doomers have been right all along. I disagree. 

These AI companies are some of the most valuable and powerful businesses to ever exist. They have the ability to warp media coverage and are responsible for the algorithms that shape our feeds. If  these corporations wanted to, they could bury these stories or spin them into a positive that paints them as heroes- but they don't. Strangely, they lean into a narrative that paints them as hapless conduits for a new world order they sincerely believe could end the human race within the near future. These stories don't seem to add up to a coercive marketing strategy at first glance, but when you remember the philosophical and economic promises these companies have used to gain enormous power, it begins to make sense. The incredible economic speculation around the AI boom is centered around the untested claim that they are going to completely and irrevocably change how our world is organized. Billions of dollars have been invested based on these claims, and investors are hoping that one of the largest gambles in history will pay off.. That is an incredibly high bar to reach. OpenAI does not have the option to merely produce a useful product that can help people schedule appointments and make Charlie Kirk memes- they are on the hook to completely alter the paradigm of human life. Investors aren't expecting a coding agent- they want The Matrix. Some of them are planning to live forever as a result of the success of AI, which has been pitched to them as inevitable and imminent. These stories serve an important role as a stopgap to keep the money flowing for these companies. It is clear from the outside that even though AI is incredibly powerful, the impact was overestimated and the timeline for AGI was at best overenthusiastic. The only way to keep up the illusion of, as Trump now calls it, ‘superintelligence’ is to manufacture a narrative that paints the internal workings of these companies as a 21st century Manhattan Project. However, when examined through a more reasonable lens, this story crumbles under its own weight. 

When David Robinson, an AI safety expert at OpenAI, resigned from his position earlier this month, it didn’t make the same waves that Coxon’s did. His reasons for leaving with a much more astute, and reasonable, warning about the lack of security measures in place (7.) In Robinson’s article for The Atlantic, he explained that a consistent lack of safety measures is indicative of larger issues within these companies that prioritize reckless progress over steady development (8.) In many ways, his claims echo that of researcher Timnit Gebru, who was fired from Google after writing her excellent paper “On Stochastic Parrots” that outlined a series of material critiques of the AI development process. The culture that Robinson, and Gebru, are referring to has long been ubiquitous throughout Silicon Valley. There is extreme pressure to continue the trend to ‘10x’, meaning a ten times improvement every 10 years. This mentality was also captured in Facebook's internal motto, ‘move fast and break things’. Under this motto, Facebook prioritized development over everything, choosing to beg for forgiveness instead of asking for permission. ‘Move Fast and Break Things’ encapsulates the neoliberal, even libertarian, worldview of Silicon Valley elites, who follow the profit incentive to its logical end, completely indifferent to the societal, economic, or environmental impacts of their behavior. In the immortal words of Maggie Thatcher, they believe that “society does not exist.” This is why, as I explained in an earlier column, Sam Altman’s OpenAI so forcefully took the reins on developing AI legislation. They wanted an environment free of pesky regulation and fully outside of the normal realm of peer-reviewed scientific development in order to maintain a competitive edge. The results of this extreme and dangerous behavior are poorly run training exercises leading to debacles like the Hugging Face attacks. The language that these stories are couched in takes agency away from these companies, furthering the fabricated narrative that the development of AI being oriented around AGI is inevitable even though our Chinese competitors, who we are supposed to be defeating, sure don't think so. These security breaches aren’t signs of superintelligence on the horizon- they are the product of a corrupt and mismanaged industry flailing violently in order to keep the dreams that they sold alive. 

So what happens if, or when, the bubble does burst? Under the current regime, most of the burden would fall on ordinary people. We already saw that SpaceX’s bloated IPO resulted in the pensions and retirements of millions being tied to the success of a historically unprofitable company. Now, both Anthropic and OpenAI are looking to start the IPO process, meaning that even more of the stock market will be tied up with AI stock. Because these companies are valued so highly, they will represent an enormous share of the market, meaning the success or failure of these companies. Outside of direct investment, the AI bubble bursting would have knock-on effects that would echo throughout the economy, resulting in an even greater transfer of wealth to the 1% (9.) Just this month, New York City comptroller Mark Levine requested that Anthropic delay their IPO until they can improve the safeguards they use to test models (10.) And why was the Comptroller of NYC so invested in their long term success? Because the pension funds that Levine controls have 7 million dollars exposed to Anthropic, a figure that would rise as soon as they went public. 

Fundamentally, these companies want us to make an incredible bet- that if we give them unlimited capital, they can eliminate all hardship from our lives. But for the vast majority of us, that trade requires giving up control of the most powerful capital we possess- our labor power. Perhaps under a more redistributive system, this would mean a universal basic income and state-owned healthcare. The downside would be handing over the last bit of power we have as workers to effect change in the world. However, I have spent my whole life watching the richest people in the world manipulate a series of economic catastrophes to consolidate wealth at the top. Why the fuck would I trust them with this?

Footnotes:

1.  “Security Incident Disclosure — July 2026.”

2.  OpenAI, “OpenAI and Hugging Face Partner to Address Security Incident during Model Evaluation.”

3.  METR and Redwood Research, Brief Independent Investigation of Agents’ Behavior, Reasoning and Collaboration in the OpenAI / Hugging Face Hacking Incident.

4.  Conger and Kim, “OpenAI’s A.I. Tried to Breach 4 Other Targets, Without Prompting.”

5.  Conger, “What to Know About Recent A.I. Hacks.”

6.  Jacob Coxon (@hilbertspaess) on X.

7.  Klein and Hu, “Opinion | ‘This Is Nuts.’ An OpenAI Insider Explains Why He Quit.”

8.  Robinson, “I Quit OpenAI Because Its Culture Is Broken.”

9.  For a more detailed description of the effect of the AI bubble bursting, I recommend: https://www.youtube.com/watch?v=wTiYaWFP59Q&t=1431s

10.  France 24, “New York City Pension Fund Asks Anthropic to Delay IPO.