Dario Amodei, co-founder and CEO of Anthropic, has been vocal about the potential dangers of artificial intelligence. He frequently warns about catastrophic risks, from autonomous weapons to AI-driven surveillance. At the same time, he has worked to position Anthropic as the company most committed to building AI safely. But this dual approach may be undermining his own goals.

In a previous interview with Fortune, Amodei stated, "AI safety continues to be the highest-level focus," arguing that this safety-first positioning has reassured corporate buyers that Anthropic's models are less risky than those of competitors. He and his sister Daniela left OpenAI in 2021 partly over disagreements about safety practices. He has also engaged with the Pentagon on concerns about autonomous weapons and AI surveillance.

Yet public skepticism is growing. A recent Gallup survey found that nearly half of adults under 30 now believe AI does more harm than good, a sharp increase from the previous year. Trust in business use of AI has fallen after two years of gains. Pew Research Center's 2026 data shows that most Americans think AI is advancing too fast, and only a small minority expect a net positive impact over the next two decades. A Quinnipiac poll revealed that more than three-quarters of Americans trust AI only rarely or sometimes, even as usage continues to climb.

In a rare lengthy post on X, Amodei wrote that he has composed one major essay on AI's risks and another on its benefits. He argued that negative public perceptions are not due to his or other AI CEOs' warnings about risks, but rather because the industry has not yet delivered on promised benefits like curing diseases. However, while trying to reassure the public, Amodei keeps associating Anthropic with "safety" in a way that implies the entire AI industry is unsafe.

This is a trap that public relations experts have recognized for decades: the problem of association. The airline industry learned long ago to never use the word "safety" in marketing. While safety reminders remain mandatory in flight briefings, airlines have transformed them into entertainment—short films, songs—rather than solemn warnings. Safety simply leaves a bad taste; it is not something people want to think about before boarding.

Historian Richard Popp's research on airline advertising shows that by the 1930s, carriers operated under unwritten rules to keep any hint of danger out of ads: no mountains, night flying, large bodies of water, or maintenance work, because each evoked the possibility of a crash. Despite this taboo, Pan Am ran a reassurance campaign in the 1980s amid hijacking and bombing threats, insisting on its safety. Then in December 1988, a bomb brought down Pan Am Flight 103 over Lockerbie, Scotland, killing 270 people. Aviation security expert Glen Winn said that disaster convinced the industry to retire the word "safe" for good.

The airline industry discovered that stating a fear about yourself is worse than saying nothing, because the word does the same work against you. Linguist George Lakoff's 2004 book "Don't Think of an Elephant" illustrates this: telling someone not to think of an elephant forces them to picture it first. The instruction to stop thinking does not erase the image.

When Amodei created a Responsible Scaling Policy modeled on biosafety tiers and issued public warnings about AI risk and job losses, it resonated with one audience: engineers, safety researchers, enterprise buyers, and investors who view Anthropic's concerns about "catastrophic risk" as diligence. But these were already likely to trust Anthropic's motives. The danger imagery did not attach to the company in the same way it might elsewhere.

Now, Amodei's direct approach may be pushing users against AI. When Anthropic aired a television ad titled "There's hope in hard questions," it opened with a burning house, then showed a crowd being scanned by facial recognition, a person sleeping on a city street, and rows of tombstones resembling a national cemetery. A voiceover asked, "Can AI be trusted?" After seeing it, OpenAI CEO Sam Altman said, "i thought this was satire," mistaking it for a parody. To a general audience, the ad did not convey that Anthropic is the only company willing to grapple honestly with AI risks. Instead, it reinforced the idea that all AI companies, including Anthropic, are ushering in a dystopian future.

That is the lesson airlines learned about selling safety. And it is one the AI industry may need to relearn today.