Virtually Open AI: How to Spin the Narrative

Addendum Sun Aug 2, 2026

The good people at NYTimes Hard Fork published an episode on the date of this post to reflect on the develops that I've spoken too. Casey Newton and Kevin Roose call people who say this is a PR stunt, completely wrong because why would a company sell a product that commits crimes?

Politely:

• Do they forget the Ultra-Rich interpret fines from illegal activities as fees to get what they want?
• Do they forget that we exist in a two-tier legal system where the Ultra-Rich can pay-to-play for their problems to go away?
• Have they so easily forgotten the lessons from Jeffery Epstein?

Shameful egotistical reporting by the NYTimes Tech Columnist Kevin Roose and Platformer Owner Casey Newton to trivialize a very real possibility of Frontier AI Labs advertising to black market players backroom deals for SLiME tools to exploit the general public. Silicon Valley's balance sheet dictates its future, and it's hungry for revenue from anyone.

Also, writing notes for a future self is an interesting perspective from the duo. Have we forgotten save-points in video games? I'm not trying to be obtuse—we must be careful with our language because SLiME tools are not sentient and do not write notes for itself. This manner of notation aggrandizes the observed feats as more than what they are: rule-set heuristics incentivizing the path of least resistance.

Original Post

On July 21 OpenAI (a Frontier AI Lab financially backed by groups such as Microsoft, Andreessen Horowitz, and Elon Musk) published a notice of a security incident which occurred sometime before July 16 on Hugging Face (a repository of published LLM models) servers. To learn more about this incident, checkout the links to their disclosures:

1. OpenAI, July 21: https://openai.com/index/hugging-face-model-evaluation-security-incident/
2. Hugging Face, July 16: https://huggingface.co/blog/security-incident-july-2026
3. Hugging Face follow-up,July 27: https://huggingface.co/blog/agent-intrusion-technical-timeline

I want to articulate three aspects of this series of events because I, an expert in Artificial Intelligence before the publication of the first ChatGPT model, have a healthy skepticism of being told what to believe by individuals in authoritative positions in the fledgling “Intelligence Marketplace.” Because of certain past correspondence with journalists around this topic, I say upfront that I am not discrediting the achievements of SLiME tools (a.k.a. modern language models) because the evidence of their feats is apparent (i.e. go read the links I provided, these are powerful tools), rather that I distrust the narratives being told by the makers of the models. People like Dario Professor Panda Amodei, Jack Clark, Sam Altman, Larry Page, Sergey Brin, Jensen Huang, Peter Thiel, Elon Musk, Marc Andreessen, Satya Nadella, Bill Gates, etc., who have positioned themselves as ultimate beneficiaries to most bifurcations of outcomes in how the world will exist with their authored definitions of SLiME tools.

The first aspect is the lexicographic construction and graphic dress of the public relations around Artificial Intelligence software. Read Hugging Face's second publication to find an interactive tool explaining this security incident: it's darkly colored, using hacker-style courier new font, with slow blinking lights, and a palette of colors and symbols that express ex-military expertise copy dress. DANGER, NUCLEAR WAR, 9/11 Part 2. This goes further with, importantly, the style of words used: reconnaissances, hunt, stage, root [it], dropper, exfiltrate, evasion, supply-chain, dead-drop, harvested, re-arms. Talk about a Command & Conquer vocabulary, any septuagenarian US Senator reading this will be quaking in their boots thinking of the 1950s Red Scare or the 1960s Cuban Missile Crisis. Meanwhile Hugging Face is labeling the Unix command ls (which means 'list the files in this directory') as reconnaissances. This facade has cracks if you look at this HTML+JS applet closer—certain key details are missing. The censor is for your protection and safety, moans OpenAI, worried that you'll somehow use their precious software in the same manner in which they do. The takeaway from this is that if information is assembled with key aspects missing, a variation on the true narrative can be told. An intellectually honest corporation would share the server log files instead of creating this glitzy applet to inspire fear.

The second aspect is the orchestration of the parties involved and the information missing from the disclosures. To continue the idea of a variant on the true narrative, we need to address the parties in the room: OpenAI, Hugging Face, … JFrog, and Microsoft. What is JFrog? They maintain a repository of published C# libraries called Artifactory, which ASP.NET developers access through a package manager called Nuget, all of which supports Microsoft's developer ecosystem for resources on Azure (the Microsoft cloud provider). Speaking from personal experience, it's a rat's nest of vulnerabilities. When reading OpenAI's notice, you will see the organization does just enough disclosure for the reader to assert “ah, there were four insecure packages on a service called Artifactory that GPT exploited,” and proceed to not disclose the packages nor authors' names. There is a promise by OpenAI that those at fault have been notified about their vulnerability and they will publish this information in due time, but will they? The takeaway here is that it is all too easy in 2026 for Microsoft, who has the ear of OpenAI (they kicked Sam Altman out before rehiring him, remember?), and control over the C# supply chain, to establish the software equivalent of Drug Mules to pony the necessary SLiME tool resources in place.

The last aspect is that there is nothing new here in SLiME tool functionality. Everything that the SLiME tool did has been available for use by models for at least a year, ever since Anthropic announced a CLI version of its SLiME tool Claude. If OpenAI is really staffed by experts who care about security, and totally not a front staffed by a bunch of naïve 20-somethings that Peter Thiel thinks look photogenic, they could have air-gapped the model servers to prevent this crisis. Literally disconnecting the server the models is on and creating their own intranet, like the US Military has done for decades.

The important questions any observer should ask themselves is Why this? And Why now? I can think of a few factors: the developing US-China standoff over China's free models threatening the profits of Silicon Valley, the growing discontent among American citizens regarding the massive data center build-out across the nation threaten the construction grift, and the growing realization on Wall Street that the cap-ex for AI buildouts makes as much sense as the SpaceX IPO. This fragile balance of codependent organizations (from Microsoft, to Nvidia, to OpenAI and Anthropic) have painted themselves into a financial corner due to their zeal of attempting to reach the touchdown zone first, so they must enact their emergency plan.

So, OpenAI's handlers needs to turn the burner up on boiling the frog of public sentiment of safety around A.I.. We are in great danger says the companies acting as The House, The Dealer, and a majority of the players. Surrender to us to protect yourselves against the harm we alone can cause. Their Greek Chorus chants, Only the regulators can save us from ourselves.

So, what's at risk, beyond everything? (That is, Andreessen and the rest of Silicon Valley's public definition of everything.) There is a forced dogma from Silicon Valley that Open Source is open borders—no security: The NYTimes continues to misrepresent Open Source software, as of the July 31st episode of The Daily (which is covering this marionette theater over Open Models in Silicon Valley), as free and visible software to mean that anyone can observe the source code and establish credibility of the expected behavior of the software. What they fail to realize again and again is that it's Ownership of Intellectual Property by the Public with accountability on the collaborators.

Software is copyrighted as a novel idea fixed to a medium, like a pattern in a rug, with novel patterns being patentable. The JavaScript algorithm for GZip is a transformation of the idea from the C source code. It may be copyrightable because of a novel addition, but there is nothing patentable. Recently Federal judge for the Northern District of California William Alsup approved a $1.5B settlement over pirated books used to train Anthropic's Claude chatbot, so long as the original copy is destroyed. One replaced the other. And there is no evidence that the new, digital copy was shown, shared, or sold outside the company. So now Anthropic, in its pursuit of compressing knowledge, is now shredding rare books. An erasion of the vestiges of our common heritage that makes Ray Bradbury's Fahrenheit 451 look neolithic.

There's going to come a point where those venture capitalists in Silicon Valley believe they own the provenance of your thoughts. That if you speak or write a certain turn of phrase, their SLiME tool will alert them with this person cannot have known that without paying a subscription to access my knowledge base because that book was destroyed 20 years before they were born. They need to be a member of the OpenAI Premium Class© to obtain a license to access those thoughts. Sounds like a Philip K. Dick novella, but these are the same group of weirdos that are funding a software venture called Nectome that promises to back up your consciousness for you to virtually exist on a mainframe.

We cannot allow our personal liberties to be bartered away as the easy solution. The correct answer is to, if this China situation is an actual problem, apply tariffs to the Chinese models in the world marketplace. But we shouldn't stoop to protectionism like this: Silicon Valley should grin and bare this challenge by letting the wave ride. Prove to us that you are the best in the world, or die on your sword like your laissez-faire mentality dictates.