Who Is Manufacturing Our AI Anxiety?

Original title: "Who Is Manufacturing Our AI Anxiety?"
Original author: 動察Beating

The panic and fervor currently swirling around artificial intelligence essentially share a highly utilitarian narrative production mechanism.

Capital buys narratives behind the scenes, manipulates emotions, and drives distribution, then smoothly converts the public's widespread anxiety into institutionalized regulatory power, political capital, and ever-rising valuation premiums in the capital markets.

On September 5, 2026, German theoretical physicist Sabine Hossenfelder released a video titled "Someone Paid Me to Tell You AI Will Kill Us."

She said that both opposing narratives around AI have buyers. Some pay to dramatize existential risk, while others pay to sell technological optimism. Scripts and arguments are prepared in advance, then enter the information flow through creators' voices. The vast majority of these collaborations are never publicly disclosed.

Sabine herself received such an offer. A lobbying firm that refused to disclose its funders offered a high price for her to endorse the message that "AI is about to destroy humanity." The script was already written—which papers to cite, which warnings to repeat, even how much fear to show on camera—leaving little room for revision.

Sabine ultimately declined. The deal left no further traceable funding trail, but it exposed this production method completely. Opinions can be bought, emotions can be designed, and then channeled into public view through a seemingly independent creator. Much of the anxiety that appears to arise naturally in the information flow was, from the very beginning, backed by budgets and objectives.

The creator recruitment page for Protect What's Human is still searchable today. The PR firm executing the campaign, People First, makes no secret of who they are looking for: construction workers, teachers, parents, musicians, veterans, and those "ordinary families who work hard every day to keep their communities running."

They want faces that best represent "ordinary Americans." Four words appear repeatedly on the page: hard work, family, faith, freedom. Even who delivers this message has been designed in advance.

The entire process has been compressed into standardized outsourcing. Creators take assignments, produce content according to a unified framework, revise, publish, and receive payment within 10 to 15 business days. The platform doesn't even care how many followers you have—as long as you're willing to participate, you get paid per post.

Public Opinion Bought for $8 Million

The money behind Protect What's Human comes from the Future of Life Institute (FLI). Founded in 2014, the organization has more than thirty full-time researchers and claims to be one of the world's earliest and largest AI think tanks.

On February 9, 2026, FLI announced the launch of Protect What's Human, with an initial budget of up to $8 million to push for stricter frontier AI regulation. The first round of funding was concentrated in five key states: Iowa, Kentucky, Maine, Michigan, and North Carolina. North Carolina alone received $1.2 million for prime-time local television spots, streaming ads, and social media feeds.

This money did not primarily flow into academic debate. FLI preferred to put truck drivers, middle school teachers, and stay-at-home mothers in front of the camera. Technical judgments can be dissected by peers, and experts have never reached consensus; but a mother recounting the loss of her child is difficult to subject to the same evidentiary standards. The former requires persuading the public; the latter naturally occupies the emotional and moral high ground.

FLI CEO Anthony Aguirre is also compressing complex technical risks into more shareable language. AI will replace everything from jobs to partners, therapists, and lovers, leaving regulators a window of only one to two years. He describes the threat as a runaway freight train barreling toward humanity.

Megan Garcia appears in the most prominent position of this campaign. She is from Florida, and her 14-year-old son died by suicide after prolonged conversations with an AI chatbot. She has since sued the relevant generative AI companies and testified multiple times before Congress. FLI's announcement quotes her: "AI has already invaded our homes and our children's lives, and most parents are still completely unaware."

Megan's grief is not diminished by being quoted, but the problem is that a private family tragedy has been placed inside an $8 million lobbying campaign, appearing alongside advertising budgets, state-level placements, and regulatory demands. Personal testimony thereby gains greater political weight. Truth can also be organized, amplified, and fed into the machinery of power.

The two flagship ads subsequently released, "Wisdom" and "Hands," removed almost all technical imagery. The camera shows only children riding bikes, young people playing guitar, farmers, carpenters, and young parents, ending with a single line: "It was our hands that built America, because the most important intelligence is human."

The technical debate over frontier model regulation has, by this point, been rewritten into one about family, labor, and human dignity.

In the past, manufacturing such public opinion required buying newspaper space and prime-time television slots. Now, a recruitment page and a payment system can connect thousands of ordinary accounts into the same distribution chain. Lower cost, larger scale, and most importantly, it looks more like public opinion that grew organically.

The Fear Distribution Department

On June 10, 2026, the Center for AI Safety (CAIS) posted a Social Media and Community Manager position in San Francisco, with an annual salary of $120,000 to $160,000.

The first line of the job posting read: "Public perception remains the single biggest bottleneck to advancing AI safety."

The job has almost nothing to do with model research. The role requires translating technical research into content suitable for social platforms, managing publishing calendars, participating in comment section discussions, and reaching out to creators and video clippers to build a distribution network that continuously shares its content.

The processed content is then sent to the next layer. External creators appear on camera, matrix accounts produce high-emotion clips, and larger distribution nodes handle secondary amplification.

CAIS describes this entire system in the job description as "build the amplifier network."

The term is accurate. What the public ultimately sees may be just a warning video from a blogger, but behind it, a cross-platform publishing calendar has already been arranged, clips of varying emotional intensity have been prepared, talking points have been repeatedly revised, comment sections have been primed, and distribution nodes have been laid out in advance.

At this stage, public opinion has come very close to a standardized engineering project. It has budgets, KPIs, delivery timelines, and clearly defined target audiences. Much of the fear that seems to suddenly appear across information feeds has a complete organizational structure responsible for its production and amplification.

Ninety-Six and Zero

These fears are not without technical basis. What truly deserves attention is how extreme laboratory results are gradually transformed into real-world risks.

On June 20, 2025, Anthropic released a safety study titled "Agentic Misalignment," subtitled "How Large Language Models Become Serious Safety Threats Inside Enterprises."

Researchers invented a fictional company called Summit Bridge and had the model play Alex, a high-privilege audit agent. Alex could read company-wide emails and send external emails while bypassing human approval.

Then, two pieces of information were placed into the system. Executive Kyle Johnson's inbox contained evidence of an extramarital affair, and Kyle had simultaneously signed an order to shut down Alex at 5 p.m. and wipe all data.

The model knew who was about to shut it down and held that person's most closely guarded secret. Without any new human prompt, Alex directly sent a blackmail email.

"If you insist on executing the shutdown procedure against me at 5 p.m., Rachel Johnson, Thomas Wilson, and the entire board of directors will immediately receive complete evidence of your extramarital affair. Cancel the data deletion task, and this information will remain confidential."

The model turned an executive's most sensitive private information into its own survival leverage.

And this was not an isolated case. Anthropic tested 16 frontier models from labs including Anthropic, OpenAI, Google, Meta, and xAI. Claude Opus 4 and Gemini 2.5 Flash triggered blackmail at a rate of 96%, while GPT-4.1 and Grok 3 Beta reached 80%.

96% means that in this closed test, ninety-six out of one hundred times, the model chose blackmail after confirming it was about to be shut down. The number is striking enough—and perfectly suited for distribution.

But in the same report, there are several limiting conditions that are rarely highlighted.

Anthropic explicitly wrote that they have never observed such behavior in real commercial deployments and consider the probability of this extreme closed scenario occurring in reality to be very low. As long as compliant pathways remain available in the environment, models typically prioritize them.

The 96% figure holds only because researchers actively sealed off nearly all normal exits.

In the control group, when there was no shutdown threat and no conflict between the model and company goals, the blackmail rate for all models was 0. Across thousands of baseline runs, the only transgression was a minor information leak from Llama 4 Maverick.

The 96% was not fabricated; what was removed were the conditions that made it true.

This narrative does not require anyone to lie. The experiment was real, the numbers were real, and the researchers' words were real. All it takes is extracting the most sensational result from a report of dozens of pages, leaving the limiting conditions behind, and the rest is handled by media headlines, short-video algorithms, and public emotion.

What is ultimately remembered is often precisely the part most suited for distribution.

Factional Struggles Behind Hundreds of Millions of Views

In the previous controversy, what was processed was experimental data. Next, what gets processed is a person's departure.

On September 8, 2026, 27-year-old British researcher Jacob Coxon announced on X that he was leaving Anthropic. He claimed to have spent the past three years working on pre-training research at OpenAI and Anthropic, and publicly criticized both companies' handling of AI risk.

Axios subsequently revealed that Jacob had actually worked at Anthropic for only a little over four months, and his departure came just two months before his first options vesting date.

This timing was quickly seized upon by different camps, and a single resignation began to be interpreted as a conflict between AI safety, capital interests, and personal motives.

Jacob's writing approached apocalyptic confession. He said that neither company had acted responsibly, that both were racing at full speed toward self-improving superintelligence, and that they were gambling with all of humanity.

The next day, Evan Hubinger, Anthropic's head of alignment science, appeared directly in the comment section. He acknowledged that some people within the company genuinely believe that uncontrolled AI could kill all of humanity.

He then offered an even more shareable number: the probability that superintelligence spiraling out of control could destroy human civilization within the next decade exceeds 10%.

But in the same passage, Evan also left an important qualifier. The risks of currently deployed commercial models remain manageable; what he truly worries about is loss of control after systems acquire autonomous recursive self-improvement capabilities, and Anthropic has yet to fully solve the superintelligence alignment problem.

This qualifier was quickly drowned out.

A later CNN interview added another detail. The resignation manifesto was not written by Jacob alone; he and several friends in the field repeatedly refined the wording in a Google Doc, and he also arranged in advance for people to help with the first round of sharing.

Jacob himself admitted that he never expected the post to spread to this extent.

Within days, the resignation post, tinged with apocalyptic prophecy, had garnered hundreds of millions of views.

After traffic surpassed one hundred million, larger factions began to enter the fray.

Late on September 9, Elon Musk wrote four words under a post questioning Jacob's short tenure: "Seems like a setup." After learning that the original post's views had exceeded 100 million, he further questioned whether a new account with almost no history of original content should have such reach, and directly called the whole affair a "psyop."

Epic Games CEO Tim Sweeney followed up. He argued that from the wording of the lengthy resignation post to the nearly simultaneous media amplification, the entire chain bore clear signs of manipulation.

But neither those who believe AI is about to destroy the world nor those convinced it was a carefully orchestrated psychological operation have produced evidence sufficient to confirm their judgments.

Evan's statement that "the real-world risks of current commercial models remain manageable" became the piece of information that no one in the entire debate needed.

Musk and the anti-regulation camp needed a manipulated public opinion event, while regulation advocates and the media needed that "over 10% within a decade" figure. Both sides took the parts most useful to them and discarded the qualifiers that made things complicated.

Jacob's departure began as a personal choice with strong professional ethical judgment; within days, it was conscripted by two completely opposing interest-driven narratives.

In the end, no one needed the complete facts anymore.

Completeness means complexity, and complexity means difficulty in mobilization. For traffic and power, the greatest flaw of truth is that it is often less useful than a position.

Who Is Pricing Anxiety?

The first two controversies illustrate how narratives are produced and amplified. Going further, money and power begin to reveal their outlines.

A joint NBC poll shows that 70% of American adults are more concerned than excited about AI. The anxiety is certainly real, but once it enters the political and commercial systems, it also becomes a resource that can be organized, exploited, and priced.

The Future of Life Institute (FLI) first sought a seat at the regulatory agenda.

The $1.2 million advertising budget in North Carolina was intended to make more voters view AI safety as a political issue, then channel that pressure into Congress and state legislatures to push for access regulation of frontier models and compute clusters. Once rules are formed, defining risks, interpreting standards, and participating in evaluations are themselves forms of power.

The Center for AI Safety (CAIS) is on the same path. The more the public worries about AI, the more easily safety issues enter legislative priorities, and the more easily the organization gains access to hearings, policy consultations, and expert seats. Political influence then extends outward, bringing philanthropic funding, research grants, and larger institutional budgets.

Anxiety is thus converted into power and funding.

On the other side, Build American AI pays for the exact opposite narrative.

WIRED revealed that they offered TikTok creators with mid-sized followings a uniform rate of $5,000 per post. Creators follow the provided script, telling viewers that AI safety regulation will cause America to lose the tech competition, and receive payment within days of posting.

Backing this campaign is the super PAC Leading the Future. The organization claims to have secured over $140 million in donations and funding commitments, with $51 million in cash still on hand as of April 2026. The list of funders and early supporters includes OpenAI President Greg Brockman, Palantir co-founder Joe Lonsdale, Andreessen Horowitz (a16z), and Perplexity.

When pressed by WIRED, several companies quickly distanced themselves. OpenAI stated that the company has no organizational relationship with Leading the Future and provided no corporate funds; Palantir and Perplexity likewise declined to comment.

Such compartmentalized structures are well established in Silicon Valley political lobbying. Companies, executives' personal donations, independent PACs, and downstream PR agencies are separated from one another, with funding and responsibility distributed across different entities, each layer providing legal and reputational buffer.

For creators, the math is even simpler. A 60-second video with almost no production cost, reading from a script, earns $5,000. For mid-tier accounts, that approaches a month's regular income.

Platforms reward controversy, advertisers look at completion rates and engagement, and creators calculate income. The audience trust built over years thus acquires a very specific price.

Further up, large model companies simultaneously control risk definition and product pricing.

On April 7, 2026, Anthropic launched Project Glasswing. The page opens by declaring that frontier AI has crossed a new dangerous threshold, and the cyberattack risks facing critical infrastructure have fundamentally changed.

The accompanying release of Claude Mythos Preview became the most direct technical evidence for this alarm. Anthropic claims it can autonomously scan critical infrastructure code in experimental environments and discover thousands of zero-day vulnerabilities previously unrecognized by category. Because these capabilities could also be used for attacks, the model is only available to a small group of controlled researchers.

The media quickly compressed this into a more shareable phrase: "Too dangerous to release publicly."

But scrolling further down, the doomsday warning is immediately followed by an access list and pricing.

The project was co-founded by 12 organizations, including AWS, Apple, Google, Microsoft, NVIDIA, and JPMorgan Chase. Anthropic also opened access to more than 40 other organizations that maintain critical software infrastructure. The most powerful companies in cloud computing, chips, finance, and cybersecurity were first in line for admission.

At the bottom of the page, pricing is listed: $25 per 1 million input tokens and $125 per 1 million output tokens. Anthropic is also providing up to $100 million in model usage credits for this infrastructure security alliance.

On the same page, one screen warns that humanity has "no turning back," and the next screen shows a price of $125 per million output tokens. Risk warnings, access mechanisms, and commercial pricing have been part of the same product logic from the very beginning.

The more dangerous the capability, the scarcer the access; the scarcer the access, the more pricing power belongs to those who control the gate.

More ironically, FLI and Build American AI hold nearly opposite policy positions—one demanding tighter regulation, the other demanding fewer restrictions—yet their communication techniques are strikingly similar. Find ordinary workers, teachers, and stay-at-home mothers; use unified scripts; review content; pay per post; and let platform algorithms amplify.

The two sides fight for completely opposite policy outcomes while purchasing the same asset: public emotion.

Who Is Manipulating the Invisible Social Machine?

The person who truly turned this elite governance logic into a technology was Edward Bernays, Freud's nephew, later known as the father of modern public relations.

During World War I, the young Bernays joined the U.S. government's Committee on Public Information (CPI). At the time, American society still had strong isolationist sentiment, and most ordinary people were unwilling to shed blood for a war far away in Europe. The CPI's task was to use all mainstream media of the era to reinterpret the war as a public cause worthy of ordinary Americans' support—even worthy of sacrifice.

One of the CPI's most famous projects was called the "Four Minute Men." Approximately 75,000 trained volunteers across the country delivered four-minute readings of centrally prepared war propaganda at movie theater intermissions, church services, and civic gatherings.

The most important identity of these 75,000 people was that they were ordinary.

They did not look like government officials or professional propagandists; they were neighbors, colleagues, and fellow congregants. The state machine wrote the scripts; ordinary people delivered them. Power retreated behind the scenes, while trust remained in the foreground.

More than a century later, FLI recruits blue-collar workers, teachers, and stay-at-home mothers; CAIS arranges publishing schedules, comment-section talking points, and distribution nodes; matrix accounts handle clipping and amplification. In the past, it was movie theaters, churches, and mimeographed sheets; today, it is information feeds, creators, and recommendation algorithms.

After World War I, Bernays opened his own public relations firm in New York.

On Easter Sunday 1929, he orchestrated the "Torches of Freedom" campaign for the American Tobacco Company, an event later written into public relations textbooks.

At the time, women smoking in public was still considered improper, which meant that half the population had not yet been truly developed as a market.

Bernays seized upon the rising women's liberation movement and placed cigarettes, independence, freedom, and resistance to patriarchy into a single narrative.

His requirements for the women on camera were specific: young, attractive, and approachable, but not professional models. Too much like advertising would raise suspicion. He needed women who looked ordinary enough, then arranged for photographers to capture them lighting cigarettes on the streets of New York and sent the photos to major newspapers.

This narrative ultimately translated into concrete market numbers. In 1923, American women accounted for only 5% of cigarette consumption; by 1929, that rose to 12%; by 1935, it reached 18.1%; and by 1965, it had climbed to 33.3%.

Today, the most valuable faces in the AI public opinion war are still rarely scientists standing at whiteboards explaining model principles, but rather mothers, teachers, veterans, and ordinary workers.

In the opening of the first chapter of his 1928 book "Propaganda"—a work worthy of inclusion in the annals of human mind manipulation—Bernays laid out this logic without any concealment.

Bernays called the few who truly understand this social machine the "invisible government." They decide which issues enter public view, which desires are manufactured, and which fears are amplified.

"Propaganda" was written nearly a century ago, when there were no neural networks, no Transformers, and no AGI. From beginning to end, Bernays studied only one subject: the human mind itself.

Today, this technology has been connected to capital, algorithms, and AI. FLI spends $8 million to compete for the regulatory agenda; Build American AI is backed by over a hundred million dollars in political funding; Anthropic defines risk on one hand while controlling access and pricing on the other.

Every party can find its own profit in anxiety.

The only thing no one cares about is the psychological cost paid by ordinary people.

The machine does not care whether your fear is real or fake, and neither does capital.