In early 2023, JPMorgan Chase banned ChatGPT. In February 2024, Klarna announced its AI assistant had handled 2.3 million customer conversations in its first month — the equivalent work of seven hundred full-time agents. In May 2025, Klarna’s CEO told Bloomberg they’d overcorrected and needed to rehire humans.
Three companies. Three decisions. Three different versions of the same organizational question: what do you do when a forcing function arrives and your refusal culture is the first thing it hits?
JPMorgan’s story is the most instructive because it contains both the refusal and the recovery. The bank banned ChatGPT alongside most of Wall Street in early 2023 — a reasonable governance response to genuine data-security concerns. But while JPMorgan deliberated, other organizations started compounding their learning. Eighteen months later, JPMorgan rolled out its own LLM Suite to over two hundred thousand employees, and President Daniel Pinto estimated the bank’s AI tools now contribute between one and 1.5 billion dollars in annual value.
That’s an impressive recovery. But the eighteen-month gap is the story. During that window, competitors were training their people, building workflows, developing institutional knowledge about where AI adds value and where it doesn’t. JPMorgan had to compress all of that learning into a fraction of the time. They succeeded — because they’re JPMorgan, with functionally unlimited resources. Most organizations can’t buy back lost time at that scale.
Samsung’s story is the pure cautionary tale. In March 2023, the company allowed ChatGPT use in its semiconductor division. Within twenty days, three separate employees uploaded proprietary data — source code, defect-detection algorithms, internal meeting transcripts — directly into the tool. Samsung responded with a company-wide ban. The security response was understandable. The opportunity cost is still accumulating.
But Klarna’s story is the richest, because it contains the lesson that neither pure refusal nor uncritical embrace gets right.
When Klarna launched its AI assistant in January 2024, the numbers were staggering: two-thirds of all customer service chats automated, resolution time cut from eleven minutes to under two, a projected forty-million-dollar profit improvement. CEO Sebastian Siemiatkowski was understandably ecstatic.
Fifteen months later, he was telling a very different story. Quality had deteriorated on complex interactions. Customer satisfaction was suffering. Siemiatkowski acknowledged publicly that cost had become too dominant a factor, and the company began rehiring human agents for high-complexity cases.
The lesson isn’t that AI failed at Klarna. The AI still handles the majority of inquiries. The lesson is that Klarna replaced one form of refusal culture — the reflexive “we can’t use AI” — with another: the reflexive “we can’t slow down.” Both are versions of refusing to make a deliberate choice.
Meanwhile, Moderna deployed over seven hundred fifty custom GPTs within months of launching ChatGPT Enterprise, with one hundred percent adoption in its legal department. Shopify CEO Tobi Lütke sent a memo declaring AI usage a baseline expectation and requiring teams to demonstrate why they can’t accomplish goals with AI before requesting additional headcount.
The pattern across all of these cases is the same one that killed Blockbuster and Kodak, just running at compressed speed. Erik Brynjolfsson’s J-Curve research explains the mechanism: general-purpose technologies demand massive complementary investments in process redesign and human capital. Organizations that refuse during the initial productivity dip — citing low measured ROI — systematically miss the downstream inflection.
BCG’s 2025 research quantifies the gap: roughly sixty percent of companies are generating no material value from AI. Not because the technology doesn’t work. Because their cultures default to “can’t” before anyone has a chance to learn what “can” looks like.
The discipline required here isn’t optimism or caution. It’s choice. Deliberate, informed, honest choice about what your organization will try, what it will protect, and what it will learn from — made transparently, not by default.
Ask your team this week: where are we already using AI informally that we haven’t sanctioned? The honest answer will tell you whether you’re governing AI adoption or just pretending it isn’t happening.