For three years, the question about AI agents in finance has been the same one: are regulators going to let software hold the pen on a settlement? Japan just answered.
On 19 May the LDP’s Digital Society Promotion Headquarters, through its Next-Generation AI and On-Chain Finance Initiative — a project team led by Seiji Kihara — approved the “Next-generation AI & Onchain Finance Concept,” a national plan that puts agentic AI, tokenised deposits and stablecoins at the centre of Japan’s payment infrastructure. Two days later the proposal was formally handed to Digital Transformation Minister Matsumoto as part of the LDP’s broader “Digital Nippon 2026” package. The Financial Services Agency has been handed a five-year clock to draft the implementing rules. The first piece of that roadmap — Payment Services Act amendments admitting foreign-issued stablecoins into Japan’s payment system — goes live on 1 June.
The framing is the news, not the timing.
The operative language in the recommendation is not hedged. The document describes a future financial system built on “automation, integration, and 24/7/365 operation, where AI agents can autonomously execute transactions, coordinate supply chains, settle payments, and manage financial activity without human intervention.”
That sentence is doing the heavy lifting. It is not “AI-assisted settlement” or “human-in-the-loop automation.” It is the affirmative case for software agents being primary actors in regulated payment flows. The plan names tokenised deposits and stablecoins as the two pillars those agents will operate on, calls for “further exploration of wholesale CBDC and tokenized Bank of Japan settlement systems,” and asks for an “AI/On-Chain Finance Asia Policy Dialogue Framework” so the rules ship with cross-border interoperability rather than getting bolted on later.
Seiji Kihara, the LDP figure most associated with the file, was quoted by crypto.news as saying “from here on, we will build it up piece by piece.” That is the right register. Nothing in the plan is self-executing. Every operative claim has to come back through the FSA as a rule.
But the political signal is the one most other jurisdictions have refused to send: agentic finance is now a state-backed infrastructure project, not a private experiment that needs to be supervised.
The PSA amendments that take effect on 1 June are narrow on paper and broad in intent. The FSA finalised the Cabinet Office Ordinance after a public-comment period running 3 February to 5 March and sixteen filed opinions; the revised ordinance reclassifies foreign trust-type stablecoins from “Securities” under the Financial Instruments and Exchange Act to “Electronic Payment Instruments” under the PSA, which means licensed Japanese intermediaries can finally hold and move them as payment assets rather than as investment products.
The qualification framework is a comparability test, not a checklist. Foreign issuers have to operate under a regime providing “protections broadly comparable to Japan’s” across four dimensions the FSA has named publicly: reserve asset management, audit obligations, redemption rights, and the supervisory authority’s ability to share information with Japanese counterparts. Stablecoins linked to elevated money-laundering or criminal risk can be rejected outright.
That structure matters for the agentic case. An AI agent settling cross-border in a foreign stablecoin needs a counterparty regulator the FSA can call. The comparability framework is the legal precondition for any agent that decides, on its own, to settle in something other than yen.
The 5-year plan does not touch the existing per-transaction and structural limits inherited from Japan’s 2022–2023 stablecoin framework. Trust-type stablecoins still pass through licensed intermediaries; bank-issued digital-money stablecoins are still treated as deposits, with holders protected up to ¥10 million by deposit insurance; the FSA’s liability-reserve regime for exchanges, scheduled to land later in 2026, is the operative consumer-protection floor.
None of that goes away because the LDP wrote a vision document. The plan even acknowledges the gap by asking — but not yet answering — whether stablecoins should become legal for wages, tax payments and capital contributions.
So the operational reality on 1 June is the same one Japanese banks were preparing for last quarter: agentic settlement is permitted in principle, gated by intermediary licensing, capped by deposit-insurance and reserve rules at the human layer, and dependent on whatever the FSA writes into the comparability assessments. The state has set the direction. The plumbing constraints are intact.
Most agentic-finance discourse outside Japan has been bottom-up: pilot at a desk, get a no-action letter, scale. The Japanese plan inverts that. The state has decided that the rails should exist, named who owns the timetable, and committed to building cross-border alignment up front.
That has three immediate consequences for anyone selling agentic settlement, payment automation or AI treasury products.
First, the addressable Japanese market is no longer just bank IT. The roadmap drags wholesale CBDC, tokenised BoJ settlement and stablecoins into the same plan, which means a single agentic stack can target all three rails over the next five years instead of pretending each is a separate vertical.
Second, “we use a foreign stablecoin” stops being an automatic compliance problem. From 1 June it becomes a comparability conversation between the issuer’s home regulator and the FSA. That is a much narrower argument to win.
Third, the Asia dialogue framework is the part most foreign vendors will under-read. Japan is the first G7 to ask, in writing, for regional alignment on agentic finance rules. If Singapore, Hong Kong and Korea respond, the agentic settlement market in Asia gets a coordination layer that the US and EU do not have. That is a meaningful structural advantage for the firms that build to Japan’s specs first.
The next test is the FSA’s draft timetable. Five years is a long horizon; the first eighteen months of consultation will reveal whether the agency intends to retrofit agentic operation onto existing PSA categories or build a parallel licensing track. The LDP plan implies the latter. The FSA’s institutional habit is the former.
The second test is whether any of the agentic language survives translation into actual regulation. “Without human intervention” is operative in the policy document. It is unlikely to survive verbatim into a PSA ordinance. The compromise wording — supervisory thresholds, human-attestation triggers, mandatory shutdown triggers — will determine how much of the original vision actually ships.
And the third test is on 1 June itself. The PSA amendments are the first concrete delivery. Watch which foreign issuers the FSA names as comparable, and which it rejects. That list will tell the agentic-finance market more about Japan’s real risk appetite than any policy paper.