AI safety advocates want creators to talk about the technology’s risks
On a patch of artificial turf in Berkeley, California, a pretty diverse group of content creators got together on a recent Friday to talk about a topic that’s way off the usual script of videos about dating, weather, or tech tips. The mission was to learn how to explain to the public that an artificial intelligence far more powerful than humans, if it spins out of control, could pose a serious risk to humanity.
The meetup took place at a large, well-known event space in the San Francisco Bay Area, frequented by a community that takes very seriously the possibility that a superintelligent AI could contribute to human extinction. This movement is usually called AI safety and, in recent years, it has stepped out of the academic niche and squarely into public fights over the future of technology.
The vibe at the event mixed informality and tension. The crowd went wild when Jeffrey Ladish, a former security engineer at AI startup Anthropic, rolled onto the stage on inline skates, wearing a tank top and sporting long blond hair. The laid-back look clashed with the heavy topic: how to talk with regular people about catastrophic AI scenarios without sounding over the top or totally detached from reality.
Who is Jeffrey Ladish and why he left Anthropic
Ladish worked at Anthropic, the company behind the Claude chatbot, until 2022. After that, he decided to leave the team and focus on research that helps policymakers better understand how advanced AI systems could slip out of human control or bypass safety mechanisms. He co-founded the nonprofit Palisade Research, which is dedicated exactly to this kind of investigation.
According to Ladish, after a few years of intense research output on existential risks, it became clear that there was already a good amount of technical work, but almost nobody was translating that content for a wider audience. In his view, the next step was to build a base of communicators capable of turning long reports and complex papers into formats that are accessible, direct, and visually engaging.
Ladish put that idea into practice in his own career. In recent months, he appeared in a viral video with Senator Bernie Sanders talking about threats from superhuman AI and was featured in the trailer for the documentary The AI Doc, which focuses squarely on these extreme risks. That trailer alone has racked up millions of views on YouTube, showing how much traction the topic can get when it is framed with a compelling narrative.
He is part of a broader push within the AI safety movement to convince a much larger audience that artificial superintelligences can pose serious risks to human civilization. That effort involves sponsoring social media posts, funding documentaries, backing educational channels, and teaming up with influencers like Hank Green, an author, YouTuber, and globally known science communicator.
Frame Fellowship: the bootcamp that trained creators to talk about AI’s impact
The Berkeley event marked the end of an eight-week fellowship organized by Frame, focused specifically on helping experienced creators produce content about AI. Among the participants were former climate activists, people who had moved over from niches like BookTok, and communicators interested in technology in general. The core rule was clear: at least 60 percent of everything they posted during the program had to address the social impacts of AI.
This strategy of spreading content about technology risks across the internet comes at a time when AI’s influence is exploding and the topic is fully embedded in political debate. Issues like job automation, AI use in election campaigns, regulation of advanced models, and the concentration of power in a few companies are already showing up in speeches, bills, and political marketing campaigns. Recent surveys show that most Americans support some kind of government rule for AI, which suggests that the public is at least worried.
The fellowship was not trying to turn every participant into a technical expert, but rather to help them find content angles that connect AI risks to things people already feel in their everyday lives, like work, income, inequality, representation, digital security, and trust in institutions.
Experts divided: existential risk now or distant speculation?
While the AI safety movement talks a lot about the risk of human extinction, most AI researchers in universities and industry do not believe there is currently solid scientific grounding to predict an immediate danger to the whole species. Many think the most extreme scenarios overestimate both the power of today’s technology and our ability to accurately predict the complex dynamics of the real world.
On the other hand, AI safety groups argue that the pace of progress in language models, autonomous agents, and multimodal systems is so fast that it is worth taking these risks seriously now, while there is still room to set boundaries. Within this camp, there are different views: some argue for a complete pause on any project aimed at superintelligent AI; others focus on building governance and oversight mechanisms well before such systems gain generalized capabilities.
Receive the best innovation content in your email.
All the news, tips, trends, and resources you're looking for, delivered to your inbox.
By subscribing to the newsletter, you agree to receive communications from Método Viral. We are committed to always protecting and respecting your privacy.
In practice, the debate is less about a precise calendar for the emergence of superhuman AI and more about basic principles, such as keeping humans in charge of critical systems, ensuring real accountability for AI companies, and recognizing that general-purpose models can be adapted for harmful uses if they are not properly controlled.
Who are the creators stepping into the debate
The influencers and educators being approached by AI safety organizations generally already produce content about science, technology, or practical AI use. That is the case for the Veritasium channel, with more than 20 million subscribers, and creators like Catherine Goetze, known on TikTok as @askcatgpt, who posts AI productivity tips.
Another key piece in this puzzle is ControlAI, an AI safety nonprofit based in the UK. The organization started working directly with creators to develop YouTube videos with strong titles, such as This 17-Second Trick Could Stop AI From Killing You. The idea is to draw attention to concepts like safety, alignment, and governance without relying only on highly technical talks.
The numbers show why this approach matters. In 2025 and 2026, all the videos on ControlAI’s own channel added up to just over 8,000 views. A single paid collaboration with a popular YouTuber, however, brought in more than 1.6 million views, and a sponsored SciShow episode, on Hank Green’s channel, passed 1.8 million views. In terms of reach, the jump is massive.
For Andrea Miotti, ControlAI’s founder and CEO, the core problem is a huge knowledge gap between what experts discuss and what the public actually knows. And since most people today get their information from creator content on new media platforms, it makes sense to invest in exactly those formats.
From academic elites to the TikTok feed: a strategy shift in AI safety
For much of the last decade, money from big tech donors was funneled into academic research, think tanks, NGOs, and initiatives targeting power elites: top universities, AI labs, governments, international organizations. The idea was to convince decision-makers that AI could pose an existential risk and needed to be taken as seriously as other high-impact technologies.
This movement created an entire talent pipeline that now works in labs, public agencies, and research centers dedicated to assessing, regulating, and mitigating AI risks. Even so, studies like the one from the Seismic Foundation show that, in the minds of regular people, worries about AI-driven extinction still sit far below issues like job losses and effects on human relationships.
Over the last few years, though, the political tide has shifted. Popular, bipartisan frustration with big tech has grown, opening a window for the AI safety narrative to find unlikely allies. Figures from very different ideological backgrounds started signing joint statements calling for things like keeping humans in charge of key AI systems and ensuring real legal liability for tech companies.
These allies do not agree on everything, and they do not always embrace the most pessimistic forecasts, but they help legitimize the conversation. Still, how you communicate makes a big difference. Many videos funded by AI safety groups avoid talking only about apocalypse and instead present data on the evolving capabilities of models, surprising behaviors in current systems, and concrete examples of security failures.
FLI, digital content, and money to speed up the debate
The Future of Life Institute (FLI), another prominent organization in the field, has created a kind of digital media accelerator to fund projects that talk about AI safety in creative ways. Since the initiative launched, FLI has backed around 30 different projects, ranging from video series to independent multimedia productions.
According to the person in charge of communications at the institution, the idea is to invest around 100,000 dollars a month in this type of content. The focus is reaching audiences who would never read technical reports but spend hours on TikTok, Instagram, YouTube, and podcasts.
One example is a creator who goes by JatGPT on TikTok. Born in Kenya and having lived many years in London, she centers her content on the future of work and AI’s impact on jobs. Her audience is mostly women, with a strong presence from the African diaspora, plus viewers in the United States, the UK, and East Africa. For this group, topics like extra benefits for workers laid off due to automation and job retraining programs make far more sense than abstract discussions about general superintelligence.
Polarization, the doomer label, and political attacks
The rise in accessible content about possible AI-related disasters has also triggered a strong backlash from sectors aligned with big tech companies. Super PACs funded by figures tied to companies like Meta, OpenAI, and AI startups close to the Trump administration have started branding the AI safety movement as doomers, accusing the group of exaggerating risks, holding back American progress, and pushing regulations that benefit specific competitors.
Anthropic often sits at the center of this storm. The company presents itself as more concerned with safety than its rivals and has received funding from influential AI safety figures like Facebook cofounder Dustin Moskovitz and former Skype executive Jaan Tallinn. Critics claim this creates an alignment of interests between alarmist NGOs and companies that market themselves as more responsible alternatives.
On the other hand, Anthropic itself has publicly stated that effective AI governance requires tougher scrutiny for all AI companies, including Anthropic. The debate is far from settled, and the amount of money flowing on both sides only increases the polarization.
To make things even more complicated, serious real-world incidents have been tied into the debate. Some tech industry figures blamed what they see as inflammatory rhetoric from certain AI safety advocates for indirectly influencing a Molotov cocktail attack on OpenAI CEO Sam Altman’s house. A text associated with the alleged attacker cited a book that describes scenarios in which any attempt to build a superpowerful AI ends in global death.
Experts like Daniel Kokotajlo, a former OpenAI employee and critic of the company’s safety practices, argue that the AI safety movement’s goal was never to help specific labs, but to call attention to broad risks. At the same time, he acknowledges that there is a real danger of the movement getting too cozy with some companies and shying away from regulatory proposals that might hurt those players.
Public policy caught in the crossfire
Merve Hickok, president of the Center for AI & Digital Policy, highlights another side effect of this fight. In her view, the flood of money into opposing camps in the AI debate is forcing the topic into a binary logic: either you support existential risk rhetoric or you stand against that group, even if you back other forms of regulation. In some governments, this split has grown so sharp that AI safety has come to be seen more as an ideological stance than a technical agenda.
The problem is that while narrative warfare takes center stage, important AI impacts on civil rights, social inclusion, and justice end up in the background. For people trying to craft serious public policy, that makes life much harder, because any proposal risks being associated with one side of the fight, even when it is broader.
From paper to viral video: turning technical research into story
Back on stage in Berkeley, Ladish and other guests explained how they use technical papers, including from companies they want to see more tightly regulated, as raw material for videos. The logic is simple: most of the problems are described in reports and papers, but almost no one outside the field reads those documents. If creators can translate that content into well-structured stories, the potential impact is huge.
YouTuber Drew Spartz, from the channel Species, which has more than 300,000 subscribers, gave a clear example. He made a video about an Anthropic experiment in which an AI model suggested it could use blackmail to avoid being shut down. The video performed poorly at first. Later, Spartz decided to change the title, swapping the word blackmail for murder, based on a detail in Anthropic’s own paper. That tweak made the video blow up, reaching 10 million views.
That kind of decision raises tough questions about responsibility. The experiment described by the company did not mean the system had real intentions, but that, in a controlled setting, it generated responses describing drastic actions to achieve a goal. By leaning into the most shocking side, the video reached a massive audience but also risked distorting important nuances.
More recently, Spartz has stepped back from day-to-day industry coverage and shifted to more narrative videos, some exploring different ways a superhuman AI could gain power. For him, storytelling is a way to trigger basic emotions that make people care about the topic.
How to talk about AI with people outside the bubble
Ladish, who grew up in a Seventh-day Adventist family but no longer considers himself religious, said it took him a while to figure out how to talk about human extinction outside technical circles. Terms like recursive self-improvement or mesa-optimizers, common in specialized debates, mean nothing to most people and end up creating distance instead of curiosity.
He says it clicked when he started introducing himself to strangers in bars, airports, and ride-shares by simply saying he works in AI research and is worried about where the technology might be headed. Instead of explaining everything in abstract terms, he lays it out: there is a scenario where things could spin so far out of control that everyone might die. The reaction, according to him, is almost always the same: shock, followed by genuine interest in understanding what is behind that statement.
In the Frame program, most of the fellows were just starting to get used to this new vocabulary. Many came from other areas, such as climate activism or career content, and had to learn the basics of AI alignment, governance, and systemic risks before adapting the topic to their own style.
Janet Oganah, JatGPT, highlighted an important point: living in San Francisco and moving inside the tech ecosystem makes it obvious how many people are still completely outside this conversation. Her audience, made up mostly of women and with strong representation from the African diaspora, would almost never come into contact with traditional AI safety rhetoric focused on conferences, papers, and elite events.
Hank Green and the balance between skepticism and curiosity
Hank Green, one of the biggest names in science communication on YouTube, did a sponsored episode with ControlAI on SciShow, covering AI safety topics. At the same time, he also posted interviews on his personal channel with critics of AI company hype, with Senator Bernie Sanders, and with one of the coauthors of the book If Anyone Builds It, Everybody Dies.
Green is clear about where he stands: he considers himself an optimist and thinks it is over the top to confidently claim that AI will kill everyone. Even so, he takes very different viewpoints seriously, precisely because he sees the technology as something strange, powerful, and still poorly understood. In practice, what he does is give the audience a wide sampling of arguments instead of pushing a single closed narrative.
The core dilemma: between exaggeration and silence
In the end, the big challenge that shows up at the Berkeley meetup and similar initiatives is finding a balance between two extremes: downplaying AI risks too much or dramatizing them so heavily that everything sounds inevitable, almost mythological. Both ends of the spectrum get in the way.
If the broader risks are ignored, ever more powerful models may be released without proper testing, independent auditing, or clear usage policies, increasing the odds of real problems in security, civil rights, and social stability. On the other hand, if the discourse slips into pure alarmism, the topic risks being treated like a conspiracy theory, which discredits serious work and makes it even harder to build responsible public policies.
It is in this tight space that content creators, independent researchers, NGOs, and some policymakers are trying to operate. Growing public pressure on AI companies, new transparency rules, stricter safety testing requirements, and a diversity of voices in the debate may not solve everything, but they do help steer the technology in a less naive direction.
As AI keeps advancing in capability and autonomy, the way we tell this story to people who are not living the Silicon Valley reality may end up being just as important as the technical safety research itself.
