Five Major Banks Just Confirmed the Problem Agentic Commerce Still Hasn't Solved
This week five banks, including the UK's NatWest, went public with joint "Agentic Commerce Principles" — and pointedly left the liability question unanswered. Read alongside new UK adoption data and a fresh Amazon-Meta standoff, it's the clearest sign yet that trust infrastructure, not payment mechanics, is what's actually missing.
Adoption is no longer a hypothetical — even in the UK
AI agents browsing and buying on UK retail sites isn't a future scenario anymore. AI-agent searches on John Lewis's platform reached 2.5% of total searches in September 2026, up from just 0.3% a year earlier — an increase of more than 800% in twelve months.
That's real, measurable, UK-specific adoption. Which makes it notable that the same week this data surfaced, five major banks went public with a joint statement about everything agentic commerce still gets wrong.
Five banks just wrote down what's missing
On 22 September 2026, five major banks — NatWest, Bank of America, ING, ASB Bank, and Capital One — published a joint set of "Agentic Commerce Principles." Between them they cover UK, US, European and Australasian retail banking, which makes this closer to an industry position than a one-off statement.
Their asks are specific:
- Mandatory disclosure whenever an AI agent is involved in a transaction
- Greater transparency over how agentic systems reach purchasing decisions
- Stronger safeguards for the customer data these systems collect and process
Notice what's not on that list: a resolved answer on liability. The banks were explicit about the gap they're leaving open — if an AI agent authorises a fraudulent purchase, it remains unclear whether the bank, the AI provider, the merchant, or the consumer bears the loss.
Five banks, across four countries, looked at the current state of agentic commerce and independently landed on the same three gaps: you don't always know an agent is involved, you can't see how it decided what to buy, and nobody has agreed who's on the hook when it gets it wrong.
Merchants are fighting the same fight, from the other side
Those aren't abstract concerns to merchants either. On 21 September, Amazon blocked Meta's new Muse shopping agent from completing purchases on Amazon.com — the third major agent Amazon has shut out, after Perplexity, Google, and OpenAI. Amazon's stated reasons: Meta didn't disclose that Muse would be accessing its platform, the agent doesn't identify itself while browsing, and it allegedly captures and stores customer credentials in the process.
Strip away the corporate framing and it's the same complaint the banks just made — lack of disclosure, no transparency into what the agent is doing, and unclear handling of sensitive data — just coming from the merchant side of the transaction instead of the bank side.
And consumers still don't trust the handoff
New research from YouGov, conducted on behalf of ACI Worldwide and published 16 September 2026, shows why both sides are nervous. Only 7% of consumers say they'd permit fully autonomous purchasing under predefined conditions. 53% are uncomfortable letting an AI assistant make purchases on their behalf at all, and 14% say they'd want to manually approve every single purchase, with no exceptions.
Adoption is climbing. Trust in the handoff to autonomy is not — at least not yet, and not without the safeguards banks, merchants and consumers are all independently asking for.
The pattern is the point
Three completely separate constituencies — banks, a major retail platform, and consumers themselves — looked at agentic commerce this month and converged on the same three requirements: disclose when an agent is involved, make its decision-making legible, and settle who is liable before something goes wrong, not after.
That's not a payments-mechanics problem. It's a trust-infrastructure problem, and it's exactly the layer Agent Pay Safe is built around: spending mandates set in advance, intent verification at the moment of purchase, and passkey approval before any money moves — so the disclosure, transparency and liability questions banks are now asking publicly have an answer built in from the start, not bolted on after the fact.
Sources: Fintech Garden, reporting on the banks' joint "Agentic Commerce Principles" and John Lewis AI-agent search data (22 Sept 2026); GeekWire, on Amazon blocking Meta's Muse agent (21 Sept 2026); YouGov/ACI Worldwide consumer research via Business Wire (16 Sept 2026).