The Washington Post recently documented a curious development in the AI safety movement: organisations that until recently spoke mostly to other researchers are now paying social-media creators to spread the message that artificial intelligence might wipe out humanity. The creators in question include people who normally make videos about romance novels, climate change or workout routines.
The reasoning, according to the safety groups, is that technical alignment papers do not get many views. The reasoning, according to most of the rest of the field, is that the safety lobby has decided to substitute volume for evidence. Both can be true at once. They probably are.
What is actually being claimed
The pitch given to influencers, summarised across the Post's reporting and follow-ups in Digital Trends and SF Standard, hovers around the idea that advanced AI might escape human control and cause catastrophic harm, possibly within a decade. Some scripts focus on job loss. Others focus on a hypothetical "rogue AI" turning on its creators. A residency programme launched in Berkeley this April is paying creators to live together while producing content on the theme.
None of this is fabricated. There are real researchers who hold versions of these concerns, including some at the leading frontier labs. The 2026 International AI Safety Report, signed by researchers from more than thirty countries, treats existential risk as a live but disputed area of inquiry, not as a confirmed prediction.
Where the messaging breaks down
The first place to look for evidence is the survey data on what AI researchers themselves believe. Most experts who do worry about superhuman AI place it decades away, not years. The Future of Life Institute and other groups note repeatedly that "in a few years" is a media framing, not a researcher consensus. The viral claim and the technical view are on different timelines.
The second issue is what current AI systems can actually do. Today's models lack consciousness, self-awareness or independent goals in any meaningful sense. They are, as the SS&C Blue Prism overview puts it, statistical pattern matchers operating over text and images. That does not mean their outputs cannot cause harm, but it means "AI hates us" is not a coherent failure mode. Confusion on this point has been a persistent feature of public AI discourse for at least five years.
The third issue is incentive structure. A creator paid to make doom content has a financial reason to make doom content. Audiences responding to doom content have a documented attentional reason to engage with it. That does not make the content false on its own, but it does make the volume of doom-tinged AI content a poor proxy for how much risk experts believe exists.
Why the framing matters
There is a real conversation about AI risk that deserves a wider audience. The arguments raised in the International AI Safety Report, and at Stanford's HAI, include serious concerns about misuse, accident, concentration of power and economic disruption. None of those concerns rest on a Skynet scenario. Most rest on much more mundane failure modes that are hard to dramatise on TikTok.
The risk of the influencer push is twofold. It dilutes the credible safety case by pairing it with sci-fi imagery, and it gives sceptics an easy reason to dismiss the entire field as theatre. As recent research on AI and conspiracy thinking suggests, the same audiences most receptive to dramatic AI claims are also the audiences hardest to walk back to a more grounded position once the dramatic version has taken root.
The honest summary is this. AI safety is a real research field with serious open questions. The paid influencer campaign is not the same thing as that field, even when individual researchers approve of it. Treat doom-flavoured viral content the way you would treat any sponsored content, and look up the underlying research before deciding what to believe.
Commentarii · 0