The End of the AI Premium

When Klarna listed in New York in September 2025, it presented itself to the market as an "AI-powered digital bank." Its first earnings call as a public company that November ran under the headline "Record-breaking Q3 as AI-Powered Digital Bank," and CEO Sebastian Siemiatkowski opened by crediting an AI-driven model working at scale.

Seven months later, at the first-quarter call in May 2026, the word AI had vanished.

Siemiatkowski and CFO Niclas Neglén spoke for around 40 mins, covering prepared remarks and analyst questions. Throughout the entire call, the company did not use the word "AI" once. The only time the subject arose was when an external analyst asked about a payments launch with Google. Handed the perfect opening, Siemiatkowski responded at length about trust, data, and the transaction layer, concluding on payment preferences without ever uttering the term.

If we take Klarna and its underlying technology as a case study for the arc of AI language, there’s insight in what this silence means for the economics of saying “AI” in public company disclosure. Klarna’s openness about its AI journey lets us examine the precise mechanism by which the term seemingly transformed from a valuation asset to something the company actively avoids. This change, clearly deliberate, suggests other public CEOs should now calculate whether mentioning AI will help or harm their share price.

Klarna's use of the word "AI" peaked in 2024. It then fell through a period of public rehiring and walk-backs, returned for the IPO listing in late 2025. It then reached zero once the shares were publicly trading by mid-2026. If the terminology were a gauge of engineering progress, use of the word AI would have likely moved in a steady upward line as capabilities were added. It would not spike for a public listing and then disappear from use, which could suggest that something beyond technology is driving this pattern.

The value of “AI” to a company often works like the value of a promise. A promise is worth the gap between what you have now and what you say you will become. At its IPO, Klarna sold the market a future bank, and the word “AI” carried that future. One way to read the disappearance, then, is that the word has simply stopped being worth anything to say. In 2024, an assistant credited with the work of 700 agents was a rare claim. By 2026, almost every company has made some version of it, so the claim commands little to no premium. For the large share of firms where AI has become a way of running the back office, which is most of the economy and is what Klarna does, the word may now have inflated close to nothing. On that reading, Klarna is an early example of a company entering a phase where a CEO’s silence on AI is the strongest indicator that the technology is finally working.

But that is only half the story. When a word loses its upside value, you might expect it to fade from headlines while still appearing in passing, a neutral term among many. What we see instead is active, disciplined avoidance. Siemiatkowski isn’t just bored by AI; he’s re-routing a direct AI question into the language of trust and data without once touching the term. That doesn't read as indifference, it reads more like caution. And it points to a second explanation: the word “AI” now carries active risk.

Between the IPO and the first-quarter call, something shifted in the liability landscape around AI language. By 2026, regulators have made it unmistakably clear that loose AI talk invites consequences. The SEC has been policing “AI washing” in public disclosures. The EU AI Act is in full enforcement, attaching compliance obligations, conformity assessments, and fines to high-risk AI systems, a category that likely covers Klarna’s credit decisions and customer interactions. In the United States, the CFPB has signalled that automated credit denials or biased outcomes will be treated as UDAAP violations, with the term “AI” acting as a red flag for plaintiff attorneys looking to assign blame to an opaque system.

In this environment, saying “AI” on an earnings call becomes a liability trigger: a single mention can invite audits, disclosure requirements, and litigation risk that didn’t exist when the term was still a marketing asset. When Siemiatkowski substitutes “trust, data, and the transaction layer,” he chooses language that is technically true, operationally descriptive, and significantly harder to weaponise in a courtroom or a regulatory filing.

So which explanation carries more weight? Maturity and liability aren’t mutually exclusive; they reinforce each other. Maturity gave Klarna the confidence to let the word go without losing competitive advantage, but liability is what likely forced the discipline. The timeline is telling. The term was used aggressively to sell the IPO narrative, then scrubbed entirely once the shares were publicly trading and the company entered a regulated disclosure environment. That pattern matches a liability-driven silence far more closely than a slow decline in marketing value. If the word had simply commoditised, you would expect it to linger in product descriptions or technical asides. Its complete absence, coupled with active deflection, suggests a legal and regulatory calculation.

The implication for other companies is this: silence on AI now conveys a dual message. It signals operational maturity and, more tellingly, legal caution. The CEO who brags about AI on a public call in 2026 may be buying a short-term perception of innovation at the cost of long-term legal exposure. Conversely, the CEO who avoids the term entirely may not be hiding a lack of capability. They may be demonstrating that they have woven AI so deeply into the business that its name has become both unnecessary and unsafe to speak.