Swisher: AI Won't End Humanity, but Its Builders Might

What happened at The Atlantic Festival

The Atlantic has published an edited interview with the tech journalist Kara Swisher, recorded at The Atlantic Festival. Atlantic executive editor Adrienne LaFrance asked the questions. The headline sets out the thesis: AI is not going to destroy humanity, but the people building it might.

The magazine's introduction describes the background. The AI industry, it says, likes to tell two stories. In the good one, superintelligent bots cure cancer and end poverty. In the bad one, they wipe out all humans. In recent weeks, the introduction adds, Silicon Valley's existential worries about the technology it is racing to build "have blown up". Swisher sees it differently. "We have to stop thinking AI is going to kill humanity," she told LaFrance. If the AI future goes sideways, she argued, bots will not be to blame. People will be. "These are people at the helm of these things."

According to the introduction, Swisher has covered Silicon Valley for several decades, first as a reporter at The Wall Street Journal and now as host of two podcasts. This is an interview, not a news report. Nearly every claim in it is Swisher's opinion, memory or prediction, and the piece was edited for length and clarity. We therefore attribute each claim to her. The piece is credited to Victoria Feng, an assistant editor at The Atlantic.

The main arguments, as Swisher stated them

People, not machines. Swisher compared AI to guns and cars. In her telling, guns do not kill people, people kill people, and a car does not choose to hit someone. Her point is that humans run these businesses and make a lot of money from them. Nuclear energy as the analogy. She said the comparison she prefers is nuclear fission: "they split the atom and, oops, it went boom." She said Oppenheimer knew what would happen, and that AI leaders know more than they admit. She asked the audience to imagine private companies building the bomb, and said AI is "even more powerful". She said tech leaders like to tell people how smart they are, and then say "we don't know" when consequences arrive. No guardrails. Swisher said there are no guardrails on these people, "ever". She said that, apart from perhaps the railroad and manufacturing booms, she cannot think of a time when ordinary people had no say over an expansion of this size. She joked that the deli on her corner has more legislation against it than most big companies. She also said tech leaders have captured politics, and she named Citizens United. She said her next book will open with Elon Musk, Sundar Pichai, Mark Zuckerberg and Jeff Bezos standing with Donald Trump at the inauguration. In her view, they learned they could "stick a coin in this president and get what you want". Why the leaders are nervous now. Swisher said she thinks they are genuinely concerned, and she gave three reasons. First, liability. She said lawsuits are piling up, like those against cigarette makers, including cases about chatbots and children. Second, cost. She said frontier AI is far more expensive than the early internet, and she predicted that today's seven frontier companies will shrink to two. Asked to bet, she said Google would be one, and that Google and Anthropic might merge. She said ChatGPT is too expensive and not sustainable. Third, the stock market. She said it depends on seven companies, and that she is amazed it has not taken a hit yet. She also called the Hugging Face hack under-covered.

The economy. LaFrance listed possible failures: a correction, swarms of bots making the internet unusable, and a bank run led by AI agents. Swisher's answer was "all of the above". Regulation and lawyers. She said Congress has not regulated tech effectively since Section 230, and that liability and negligence cases are gaining ground, much as with cigarettes. She said the country never had privacy legislation or algorithmic transparency, and that there are about a dozen steps that would not hurt their businesses. She called anger over data centers a proxy for wider anger about what the technology does to people's lives. China. She rejected what she calls the "Xi or me argument", the claim that the choice is China or the United States. She said global cooperation, especially with China, is needed. She also said China has more strictures on AI than the United States does. What to do. Swisher said not to buy gold bars, and not to join the leaders' apocalypse plans. She urged people to go to their legislatures and to keep pushing back on data centers. "The only people I believe in in this country are voters," she said.

Background and why it matters

The claim that people, not machines, carry the moral weight of AI is not new. Swisher's version is notable for its target. She is not arguing that AI is harmless. In the same interview she said AI "can set off a nuclear bomb" and that "humans with superhuman intelligence are going to do bad things". Her argument is about where responsibility sits.

Two general ideas help readers place her remarks. Product liability is the body of law that lets injured people sue makers of dangerous products. Section 230, which LaFrance explained in the interview, is the law that keeps platform owners from being treated as publishers of their users' material. In Swisher's account, platforms "can't be sued. Now they are": liability cases are starting to press against the shelter that Section 230 gave.

Our analysis

Our reading is that Swisher moves the debate from a speculative question, will machines turn on us, to a present one: who is accountable, and to whom. That is a practical shift. Courts, state legislatures and voters can act on accountability today. They cannot act on a hypothetical superintelligence.

Her argument has an internal tension, and it is worth naming. She says AI will not kill humanity, yet she also says it is more powerful than nuclear energy and that any of the economic disaster scenarios are possible. We read this as a claim about causes, not about severity. Harm is possible and could be large, she says, but it would come from human decisions.

Her explanation of why leaders are calling for caution now, liability, cost and market fragility, is an interpretation of motive. It may be right, but the interview offers no evidence from the leaders themselves. It is best read as a hypothesis from a long-time observer.

Limits of the source and open questions

• The source is an edited interview. Words are Swisher's, and the edits are not shown.
• Her predictions are predictions. The merger of two frontier firms, a stock-market hit and the arrival of a wave of lawsuits are things she expects, not things the piece shows.
• Some figures look uncertain. She cited a Facebook settlement of $17 billion and contrasted it with a last fine of "$5 million".

The second figure may be a transcription slip, and the piece gives no source for either. She also said "10 frontier companies" and then corrected to seven. At the end she said "a 10 percent chance" and then "the percent is actually 14", with no context.

• She makes sharp personal remarks about named executives. The piece includes no reply from them, so we do not repeat those remarks as fact.
• The Atlantic's introduction does not give the date of the festival session.

Practical takeaways

For developers and product teams: treat liability as a design input. Swisher's warning is that safety shortcuts may come back as lawsuits, especially for products that reach children.

For policy watchers: she points to state-level activity, product-liability cases and local fights over data centers as the places where pressure is building. She also names privacy law and algorithmic transparency as steps she says would not hurt these businesses.

For general readers: treat statements from AI executives, including calls for regulation, as statements from interested parties. LaFrance raised the same doubt in the interview, when she described Mark Zuckerberg's earlier "Regulate me, bro, please" posture with some cynicism. Treat her own remarks the same way: they are an experienced journalist's opinions, worth weighing, not verified facts.