The Strange Marketing Genius of “AI Doom”
There is something genuinely strange about the way most artificial intelligence companies talk about their own products. Traditionally, most companies try to convince customers that their technology is useful, reliable, and safe. Frontier AI companies have often done something more unusual: they have warned that their systems may become extraordinarily powerful, eliminate large categories of work, enable catastrophic misuse, or eventually become difficult or impossible to control.
Those warnings may be sincere, but sincerity does not eliminate incentives to spread those narratives. At their core, these are barely disguised marketing claims: If this technology may become too powerful for society to safely control, then it must be extraordinarily powerful.
This dynamic is especially potent because AI companies are not merely competing on price, benchmarks, or product features. They are competing over the narrative describing what category of technology they are building. If they position AI as just another useful software tool, frontier model companies are just another software vendor. If AI can reorganize labor markets, alter geopolitical power, and create civilizational risk, the companies building it become strategic institutions that demand attention, capital, political relevance, and market influence.
In Arthashastra, Kauṭilya is not interested in taking stated intentions at face value, focusing instead on incentives, power, alliances, material interests, and the structure of advantage. The useful question is not simply whether AI executives sincerely believe their own safety warnings. The better question is what happens to the distribution of power if everyone else believes them too.
Suppose governments accept that frontier AI creates exceptional risks requiring exceptional oversight. The public-interest case for regulation may be completely legitimate while the resulting structure also benefits the largest incumbents.
Bill Gurley has made a related argument through the lens of regulatory capture: incumbents often help shape regulatory systems they are uniquely equipped to satisfy to the detriment of potential competitors. In AI, a regime requiring extensive evaluations, cybersecurity programs, specialized safety teams, regulatory reporting, outside audits, and sophisticated legal infrastructure may genuinely reduce risk. It is also far easier for a frontier lab with billions of dollars in capital to comply with than for a small startup or open-source team. A regulation can improve safety and raise barriers to entry at the same time.
That is why arguments about bad faith are mostly beside the point. Regulatory capture does not require a conspiracy. Institutions naturally advocate for solutions compatible with their capabilities, interests, and worldview. Over time, public benefit and institutional advantage become difficult to separate.
The same logic extends beyond regulation. AI safety has become an economic ecosystem in its own right—spanning researchers, governance specialists, red teams, auditors, lawyers, and compliance functions. The rise of profit-oriented companies solving problems does not mean the risks are fake; cybersecurity is a large industry because cyberattacks are real. But once a problem creates a market around solving it, that market develops its own incentives.
Fear itself can therefore have economic value. The more consequential AI is perceived to be, the greater the demand for institutions that can evaluate, secure, certfiy, govern, and regulate it. At the same time, repeated warnings about extraordinary future capabilities grab headlines and strengthen the broader commercial narrative around the technology. Current shortcomings become easier to interpret as temporary if investors, policymakers, and customers believe the long-term destination is superintelligence. Enormous capital expenditures, extreme valuations, and intense talent competition all make more sense inside a story in which AI is a technology likely to transform civilization.
Silicon Valley has always sold the future, but AI doom creates an unusually simplified version of that tradition. If AI produces unimaginable abundance, the technology is important. If it radically restructures the labor market, the technology is important. If it creates existential risk, paying attention is critical. The disagreement shifts from whether the technology matters to what form its importance will take and who to trust.
None of this tells us how dangerous AI actually is; that remains a technical and empirical question. It does suggest that we should resist treating the public conversation around AI safety as though it exists outside ordinary economics. Frontier labs, governments, safety researchers, investors, and critics all have incentives.
The Kauṭilyan lesson is not that everyone is lying. It is that people can sincerely believe something that also increases their power, institutions can pursue legitimate objectives through arrangements that strengthen their position, and a real risk can produce a market whose participants benefit from taking that risk seriously.
Artificial intelligence did not invent any of these dynamics. It has simply produced another version of them. Most marketing tries to persuade you that a product is better than you think. The strange marketing genius of AI doom is that the warning and the advertisement can be exactly the same sentence.
