Silicon Valley AI Accelerationists Fight Doomer Mind Virus
Left 0%
Center 50%
Right 50%
0 left · 1 center · 1 right
What happened
Major AI labs have reported that training of new models is becoming more automated: Anthropic said AI now leads about 26 percent of its research-and-development work and collaborates on 90 percent, and an OpenAI employee told The Information that the company has largely automated training of experimental models. Anthropic CEO Dario Amodei published “We Must Pace the Frontier,” urging firms to slow their most powerful systems, embed outside evaluators, and work with government on regulation; President Trump and senior officials rejected new federal guardrails and said companies could regulate themselves. News publishers including Mother Jones, The New York Times, and The Intercept have sued OpenAI and Microsoft over training on copyrighted articles, and a plaintiffs’ brief filed September 4 quoted internal company documents on scraping news content and substituting for publishers’ labor. Physicist Guillaume Verdon, founder of the Effective Accelerationism movement, publicly rejected claims that AI will kill humanity and attacked what he described as a doomer “mind virus” in the big labs, while a new analysis found AI feedback loops have not yet met recursive-self-improvement benchmarks; The Wall Street Journal reported that independent researchers used Anthropic’s Claude to break into OpenAI in July. A Pew Research Center survey of 3,488 adults in June found 52 percent were more concerned than excited about AI’s growing role in daily life.
Sourcing check · Fully on the record · 5 articles
4 of 4 articles cite a named source for the story. It was on the record from the first report.
What these terms mean and how they're counted
- On the record · names a source 4 of 4
- On the record means the article cites a source by name for the story's main claim: a person, or an institution speaking officially, such as a company statement or a named agency's spokesperson. The opposite is an unnamed source. The count is how many of the counted articles cite at least one such source; the names are everyone named across them, each once however the articles spell it. Here: Guillaume Verdon, Dario Amodei, Anthropic, Brent Hecht, Jack Clark and Leah Siskind
- Primary evidence · cites a document or data 1 of 4
- Primary evidence means the article cites a record you could check yourself: a court filing or ruling, a dataset or report, a recording or transcript, an official statement or press release, market data. A quote is not a document. The count is how many articles cite one. Here: court filing, internal document
- Credits another outlet 0 of 4
- The article says the story comes from another outlet's reporting (“first reported by Reuters”). Those articles relay the story; they are not independent confirmation. Here: Each article presents its own reporting.
- Unnamed sources 1 of 4
- The main claim is attributed to people who are not identified: “people familiar with the matter”, “officials said”. The count is how many articles do this.
- How it developed
- When the first report landed and whether it named anyone; when the first named source and the first document appeared. Here: First report Sep 21, 23:01 UTC by New York Post, naming its source. First document 12 hours later, from Mother Jones (court filing).
- Not counted 1
- Opinion, analysis and video pieces, and articles too short to assess. Every count above is out of the counted articles. An article republished under a new headline is one article.
This is about who is cited, not whether the story is true. Each source below says what it names. How we count
Omitted — what each side leaves out
Unpacked
Left-leaning and right-leaning coverage are not arguing the same AI-risk story. Left-leaning coverage asks whether firms that trained on others' work can be trusted with safety. Right-leaning coverage asks whether extinction warnings are a bid to lock in the biggest labs. The accelerationist charge that new rules would freeze out rivals is absent from left-leaning coverage. The publishers' lawsuit and staff warnings that training was wrecking the news supply are absent from right-leaning coverage.
The labels do not overlap. Left-leaning coverage repeats staff language calling training "the largest theft of labor in human history." Right-leaning coverage repeats the accelerationist name for safety culture, a "doomer mind virus."
How much of the next model is already being built by the current one?
Bottom line
Mother Jones never mentions accelerationists calling safety auditors a "cartel." The New York Post never mentions Microsoft staff warning that training would wreck the news supply the models need.
The receipts — 2 facts one side's coverage lacks
- OpenAI added a filter after being sued that blocked output from plaintiff publishers only, which the brief says was to stop evidence-gathering rather than to stop infringement
- Microsoft research found 90 percent fewer clicks to news articles from AI search than from traditional search
The Left View
Mother Jones argues that the same executives now asking to keep building—while warning their systems “could kill us all within a few years”—have already shown they will not police themselves. Drawing on the publishers’ lawsuit, it highlights internal language that millions would soon see models “hoovering up” their work as “an astonishing theft of unprecedented proportions,” and it attributes to Microsoft applied-science director Brent Hecht the charge of perhaps “the largest theft of labor in human history.” The piece says firms knew they were substituting for cultural labor, collapsing referral traffic (Microsoft research found 90 percent fewer clicks to news from AI search), and starting a “doom loop” that would “threaten the economic foundations of its essential suppliers,” yet still stripped ownership signals from training data and, in Greg Brockman’s case, replied “ah nice” to a “hack to get around nytimes paywall.” After suits were filed, OpenAI filtered outputs from plaintiff publishers not to stop infringement but “to stop Plaintiffs from gathering evidence,” which Hecht warned could become an “accidental cover up.” Mother Jones maps that prisoner’s dilemma onto safety: the industry would benefit if everyone paid—or slowed down—but “each individual company is better off taking content for free while others pay,” so labs “plow ahead while waiting for others to slow down.” Its conclusion is that the public cannot leave “stealing content” or “the fate of humanity” to those executives, who “will make the wrong choices—even when they know those choices are bad for them, too.”
The Right View
Accelerationists in the New York Post treat extinction warnings from Anthropic and OpenAI as overstated ideology, not demonstrated engineering. Verdon said, “I don’t think AI is going to kill us all. I think that sort of thinking is more dangerous than the AI itself,” and added that doomers “can never flesh out all the details”: “I’m a mathematician. Show me the math.” He argues Effective Altruism has “culturally, ideologically compromised all these big labs,” that proposed third-party auditors are “all friends” in a “revolving door,” and that the result would be “a sort of cartel with fake checks and balances” protecting firms that already spent billions. In that frame, Amodei’s call to pace the frontier is less a conversion on safety than a bid to freeze smaller and open-source rivals; Verdon wants more competition, more open models, and “freedom to build,” and notes the Trump administration’s rejection of new guardrails. The Washington Examiner recapped standard catastrophe scenarios—an agent releasing a virus, reaching nuclear weapons, or crippling the internet and banking—while billing the “AI apocalypse hypothesis” as having “a fatal flaw: The training data.” Data scientist Bojan Tunguz, calling himself only “adjacent” to effective accelerationism, rejected “acceleration at any cost,” said “some pacing might be justifiable,” and argued “we all have our own mind viruses,” with the live question being “just how far.”
Our Take (balanced)
The left’s strongest argument is about revealed incentives: contemporaneous internal documents in which Microsoft and OpenAI staff describe unconsented training as theft, measure the collapse of clicks to human publishers, and keep going because defection pays, are hard evidence that labs will not voluntarily absorb collective harms—and Mother Jones treats existential risk as the same race. The right’s strongest argument is that the labs now demanding a slower, audited frontier have not produced a mechanistic, testable path to human extinction (Verdon’s “show me the math”), while a published analysis finds recursive self-improvement is not here yet, which weakens treating today’s coding automation as proof of imminent loss of control. The unresolved tension is whether insider alarms plus the copyright record show racing firms know they are building systems they cannot govern, or whether those alarms are an incumbent strategy to regulate competitors while the same companies continue to scale.
4 sources
More stories
The week's bottom lines, in your inbox
One email a week: the five stories that mattered and what they actually mean. Free.