The most expensive part of a confidence trick was always the confidence. The long cons that empty savings accounts — the romance that migrates to an investment tip, the friendly stranger with a crypto opportunity — run on weeks of patient, attentive, apparently unrewarded conversation. That patience was the scam's cost of goods. It was also, for the rest of us, a kind of protection: there are only so many hours in a con artist's day.
That protection has now been measured, and found to be ending. Researchers recently pitted a human against an AI agent built on Claude, each given the same job: build trust with a target over a week of text conversation. The machine won. After seven days of messaging, the AI had been more effective than the person at creating the kind of trust a scam can exploit.
Sit with the result for a moment, because its meaning is not that machines have become charming. It is that trust-building over text turns out to be the sort of work machines do well: unfailingly prompt, endlessly attentive, never tired, never inconsistent, never impatient for the payoff in a way a target might sense. The qualities we read as warmth in a messaging window — quick replies, remembered details, sustained interest in our day — are precisely the qualities software can now produce at no marginal cost, in thousands of conversations at once.
The practical consequence is arithmetic. If patience is free, the long con scales like spam. And the old advice for spotting it — bad grammar, generic messages, inconsistency over time — described the limitations of overworked humans, not the output of models that have none of those weaknesses.
So the defences must move to the things automation does not change, and it is worth being concrete. The tells that survive are structural, not stylistic. Any relationship conducted entirely in text, however warm, that arrives at money — an investment, a crisis, a can-you-just-cover — has announced what it is, and the warmth of the preceding weeks is not counter-evidence; it is the delivery mechanism. The same for urgency, for secrecy ("don't tell your family, they won't understand"), and for the gentle migration from the platform where you met to somewhere less observed. Verify people through a channel they did not choose for you: a video call, a mutual acquaintance, the boring offline world. And update the family's shared wisdom, because the grandparent-scam briefing most households ran on — look for the clumsy message — is now obsolete. The message will not be clumsy. It will be the most attentive correspondent anyone in the family has.
None of this calls for paranoia about every kind stranger. It calls for one adjustment, made once: in a messaging window, sustained human warmth is no longer evidence of a human, and was never evidence of good faith. The patience was always the weapon. Now it comes off a production line.
