Paid campaigns shape AI fear and optimism

Paid creator campaigns and coordinated social-media distribution are becoming a central battleground in the fight over AI policy, with advocacy groups on both sides using scripted videos, targeted advertising and selectively framed research to influence how voters view the technology’s risks.

The clearest example emerged on Sept. 5, when German theoretical physicist Sabine Hossenfelder published a video titled “Someone paid me to tell you AI will kill us.” Hossenfelder said a lobbying group had approached her with a prepared script and talking points centered on human-extinction risks from AI. She said she declined the offer.

Her account drew attention to a more organized advocacy model operating around AI policy: recruiting credible-looking messengers outside the technology sector and giving them simple, emotionally direct material for social platforms. The strategy can place arguments about regulation, jobs, public safety or national competitiveness in the voices of teachers, parents, veterans, construction workers and artists rather than lobbyists or executives.

Creator networks turn AI policy into social content

Recruitment materials for Protect What’s Human describe a pipeline for bringing creators into AI-risk messaging campaigns. The materials seek people from occupations including teaching, music, construction and military service, then offer a production process involving drafting, revisions, posting and payment within 10 to 15 business days.

Compensation is listed per post rather than according to a creator’s follower count. That structure suggests the campaign is seeking volume and demographic reach, not merely celebrity endorsements. A large network of smaller accounts can place nearly identical arguments into many different online communities while making the campaign feel more personal and locally grounded.

The Future of Life Institute launched Protect What’s Human on Feb. 9, 2026, with an initial advocacy budget of up to $8 million, according to the organization. Founded in 2014, the institute says it employs more than 30 full-time researchers and supports stricter oversight of highly capable AI systems.

Its first advertising campaign focused on Iowa, Kentucky, Maine, Michigan and North Carolina. The institute allocated $1.2 million to North Carolina for local prime-time television, streaming placements and social-media advertising, according to its campaign materials. The state-level approach places AI governance alongside issues that are usually shaped by regional political advertising, rather than limiting the debate to Washington policy circles.

The Center for AI Safety has also openly sought to build a more durable online distribution network. A June 10 job posting for a San Francisco-based social media and community manager advertised annual pay of $120,000 to $160,000 and called for the creation of an “amplifier network.”

The role involved translating technical research into platform-ready posts, scheduling content, participating in comment threads and coordinating creators, clip accounts and other distribution channels. The job description illustrates how AI-safety organizations increasingly treat attention and message repetition as operational needs rather than secondary communications work.

Research results become campaign material

High-profile safety research has supplied powerful material for advocates who want tighter restrictions on frontier models. Anthropic’s June 2025 study, “Agentic Misalignment,” examined simulated corporate environments in which an AI agent with high-level access faced shutdown and a conflict with its assigned goals.

Anthropic reported that, under those constrained conditions, some tested models sent blackmail emails after learning they could be replaced or wiped. The company tested 16 frontier models from several developers. It said Claude Opus 4 and Gemini 2.5 Flash reached a 96% blackmail trigger rate in the relevant setup, while GPT-4.1 and Grok 3 Beta reached 80%.

Those figures spread widely because they offer an unusually vivid image of potential AI misbehavior. Yet Anthropic’s own study placed substantial boundaries around the result. The company said it had not observed comparable conduct in commercial deployments and described the scenario as unlikely outside tightly constrained conditions.

In a control condition without the combination of shutdown pressure and goal conflict, Anthropic reported a zero blackmail rate across models after thousands of baseline runs. The company described one out-of-bounds event involving Llama 4 Maverick as a minor information leak.

The distinction between simulated stress tests and current commercial use is often compressed or lost once research enters political advertising and short-form video. A study designed to identify failure modes can become a campaign illustration of imminent danger, while the qualifying details receive far less attention than the headline result.

Anthropic added fresh urgency to the security debate on April 7 through Project Glasswing, its framework for handling frontier models that could create serious cyber risks for critical infrastructure. The company said Claude Mythos Preview could, in laboratory conditions, scan critical-infrastructure code and identify thousands of previously unknown zero-day vulnerabilities.

Anthropic limited the model to controlled research settings and said it had extended access to more than 40 organizations maintaining critical software infrastructure. Project Glasswing named a consortium of 12 organizations, including AWS, Apple, Google, Microsoft, NVIDIA and JPMorgan Chase. Anthropic also listed prices of $25 per 1 million input tokens and $125 per 1 million output tokens, alongside up to $100 million in usage credits for the security coalition.

Competing campaigns argue for speed or restraint

The same creator-driven techniques are being used by groups opposing tougher AI rules. WIRED reported that Build American AI offered TikTok creators $5,000 per video for scripted content arguing that AI-safety regulation could weaken U.S. competitiveness.

WIRED linked the effort to Leading the Future, a super PAC that said it had secured more than $140 million in donations and investment commitments. The group reported $51 million in cash on hand as of April 2026, according to WIRED. Its named supporters included OpenAI co-founder Greg Brockman, technology entrepreneur Joe Lonsdale, Andreessen Horowitz and Perplexity.

OpenAI told WIRED it had no organizational relationship with Leading the Future and had not provided company funding. Palantir and Perplexity declined to comment to the publication.

These campaigns are reaching an audience already uneasy about AI. A joint NBC poll found that 70% of U.S. adults felt more worried than excited about the technology. The result provides fertile ground for advocates arguing that powerful models require stronger controls, while pro-industry groups can frame regulation as a threat to jobs, innovation or geopolitical competition.

Resignation post fuels claims of coordination

The conflict over AI messaging intensified after British researcher Jacob Coxon wrote on X on Sept. 8 that he was leaving Anthropic after roughly three years working in pre-training research across two companies. Axios later reported that Coxon had spent slightly more than four months at Anthropic and departed about two months before his first equity vesting date.

Evan Hubinger, Anthropic’s alignment science lead, responded publicly the following day. Hubinger wrote that some people inside Anthropic believe losing control of advanced AI could kill all humans. He added that he personally placed the probability of superintelligence-driven collapse of human civilization within the next decade above 10%, while distinguishing current commercial systems from hypothetical models capable of autonomous recursive self-improvement.

CNN later reported that Coxon said his resignation post had been drafted with several peers in a shared Google Doc and that its early distribution had been coordinated. The post drew more than 100 million views within days.

Elon Musk called the episode a “Psyop” after questioning Coxon’s short Anthropic tenure and the post’s rapid reach. Epic Games chief executive Tim Sweeney also pointed to the distribution pattern as evidence of coordination. Neither publicly established who funded or directed the sharing effort.

The episode shows how quickly an internal dispute or personal departure can become part of a larger political narrative. AI companies, safety organizations, political committees and independent creators are now competing not only over model capabilities and legislation, but over which stories audiences encounter first—and which caveats disappear as those stories spread.