How fast do users trust AI agents with payment data?

On the Invest Like the Best podcast, Instinct's CEO gave figures on how quickly users hand sensitive data to the startup's personal AI shopping and travel agent: after three weeks, "40% of the user base shares the credit card with the Instinct AI agent." He also reported that once a user shares even one piece of sensitive information, retention climbs to 80%, which the excerpt's author, Rihard Jarc, calls "very high for a consumer product." The same account puts $1 billion of transaction value already flowing through Instinct's agent, with travel use cases making up 50% of that volume.

The reasoning offered is behavioral rather than a formal model: the CEO treats card-sharing and the disclosure of "sensitive information" as proxies for trust, where the willingness to hand an agent payment credentials marks a threshold that then locks in habitual use. The $1 billion figure, with half in travel, is offered as evidence that this is not a stated-preference curiosity but is already converting into real transaction volume, concentrated in exactly the travel category the broader piece treats as an early battleground for agent commerce.

This adds a direct-adoption data point to Do consumers actually want to delegate shopping decisions to AI?: Accenture's survey finds stated trust running ahead of actual delegation, while Instinct's CEO reports behavioral follow-through — card sharing, retention — inside one named product. It also sits in tension with the caution in Can an AI agent serve both merchant and user interests fairly?: Azhar's concern is that referral-fee incentives compromise an agent's loyalty, yet the Instinct figures suggest users are sharing payment data and staying retained regardless of whether such conflicts are visible to them. Are AI agents making attention-based ads irrelevant for aggregators?, drawn from the same source, describes the aggregator side of this same shift; this note describes the consumer-adoption side.

The figures are self-reported by a single startup's CEO on a podcast, not independently audited or benchmarked against a defined cohort, so the 40% and 80% rates cannot be read as representative of agent commerce generally, only as one vendor's account of its own early users. The excerpt does not say how "sensitive information" was defined beyond the card-sharing example, nor whether the $1 billion figure is cumulative since launch or over some stated period. If the pattern holds beyond this one product, it suggests the usual trust barrier to agent commerce is lower, and crossed faster, than survey-based trust measures alone would predict.