Plaid guarantees the transfer. Hive proves the agent had authority to send it.
On May 19, 2026, Plaid launched Guaranteed Payments. If Plaid approves an ACH transfer and it later fails, Plaid covers the loss. Approval rates run as high as 90 percent, backed by Plaid Signal ($230B in transactions scored across 12,000+ financial institutions) and Plaid Protect (500M+ accounts, 1B+ devices). On May 18, the new Ti3 fraud model caught 41 percent more fraud at the same false-alarm rate. On May 21, Plaid shipped two foundational AI models for transaction classification. ACH just became a real-time rail. Here is the piece nobody has shipped yet: when the originator is an AI agent, who proves the agent was allowed to send that transfer, to that counterparty, at that amount, under that policy? Hive built that piece.
Why now
May 18, 2026: Plaid debuted Ti3, a fraud-detection framework that identifies coordinated fraud across devices, user profiles, financial institutions, and digital identities in a single pass. Catches up to 41 percent more fraud than previous iterations at the same false-alarm rate.
May 19, 2026: Plaid launched Guaranteed Payments. Plaid evaluates ACH transactions in real time. If Plaid approves the transfer, the partner app credits the user instantly. If the payment later fails, Plaid covers the loss. Built on Signal ($230B scored, 12K+ institutions) and Protect (500M+ accounts, 1B+ devices). Early customers include Kalshi and Polymarket.
May 21, 2026: Plaid introduced two in-house foundational AI models: one for instant transaction-level categorization, one for tracking behavior patterns over time. Primary categorization accuracy is up 13 percent. Loan payment detection is up 14 percent. Income classification runs at 89 percent precision.
Plaid's CTO publicly named the next markets for Guaranteed Payments: peer-to-peer platforms, digital brokerages, and cryptocurrency exchanges. Every one of those markets is building AI agents that will send transfers on their own. ACH just got fast and backed by real money. What's still missing: who signs off that the agent had the authority to send the transfer in the first place?
Angle 1 Plaid signs payment risk. Hive signs agent authorization.
Plaid's signature on an approved ACH transfer says: this account, this amount, this counterparty pattern, this device, this user behavior, we will stand behind it. That signature is the missing piece for real-time ACH. But by design, it does not answer who told the agent to send the transfer. When the sender is a software agent acting for a corporate treasury, a brokerage client, or a P2P user, the agent needs its own check before it sends. Six separate gates run locally before the ACH transfer goes out: daily cap, jurisdiction, sanctions, counterparty allowlist, KYC tier, position limit. All six have to pass, or the transfer never happens.
What every Plaid-backed agent stack gets
- Per-spend SHOD gates evaluated locally by the agent. The transfer call to Plaid never happens unless all six pass. The agent cannot exceed its hire-time scope ceiling.
- Dual-signed HAHS scope receipt persisted on Base 8453. Ed25519 plus ML-DSA-65. Verifiable by Plaid, by the customer, by the regulator, all from the same canonical record.
- No change to Plaid's risk model. Hive sits upstream of Plaid Signal and Protect. The agent's authorization receipt rides along with the transfer call as an optional header or sidecar log line. Plaid's network signal pipeline stays the same.
- Hire-time and chain-time policy separation. When the agent's mandate changes mid-task (a new spending cap, a new counterparty list), the receipt records both the original limit and the current scope at the moment it was issued. The audit trail survives the policy change.
Angle 2 Two signatures, one ledger
The CFO of a brokerage running an agentic treasury workflow opens one report. For every transfer the firm's agent sent, that report shows which policy allowed it (Hive HAHS receipt), which gates passed (Hive SHOD attestation), and what Plaid signed off on (Guaranteed Payments approval ID with the liability flag). Three pieces of evidence, one record. Nobody has to just trust the agent. Anyone can check the chain of signatures from end to end.
Net effect
- Plaid's signature: "We approved this ACH, we stand behind it, here is the network signal that justified the decision."
- Hive SHOD attestation: "All six outbound gates passed for this agent at this moment, under this hire-time policy."
- Hive HAHS receipt: "The agent had authority under policy
pol_treasury_v2.1for this counterparty class, this amount range, this jurisdiction, within this TTL window." - What the regulator reads: one view showing all three signatures, with no strategy details, counterparty personal data, or model internals exposed.
Angle 3 ViewKey: same record, four lenses
Inside a brokerage, a P2P platform, or a crypto exchange running agentic ACH transfers, four different people need to read the receipt. The CFO needs spend totals and vendor breakdowns. The compliance team needs the gate results and the agent's identity. The auditor needs to see the policy in effect at the moment of the transfer. The regulator needs the attestation chain without the strategy behind it. Today, the operations team puts these four reports together by hand. ViewKey delivers one signed file with four separate keys. Each person unlocks only their own slice.
The operations team stops running a compliance report factory. Each person pulls their own view from the same signed receipt. Instead of reconciling reports, they just check the receipt.
Architecture Where Hive plugs in
- Agent layer: before the transfer call goes out, the agent runs SHOD checks locally. If all six gates pass, Hive signs a HAHS scope receipt and anchors it on Base 8453.
- Plaid Guaranteed Payments call: the agent attaches the HAHS receipt ID as a sidecar log line or optional metadata field. Plaid's risk pipeline doesn't change.
- Plaid signs the ACH approval: standard flow. The Guaranteed Payments approval ID gets recorded next to the Hive receipt in the firm's audit ledger.
- CFO, compliance, audit, and regulator: each one unlocks their own ViewKey view from the same signed file. No export pipeline. No manual reconciliation.
Two ways to start
For fintech builders shipping agentic ACH transfers: a one-week sandbox where Hive's SHOD attestation engine sits in front of a test agent sending ACH transfers through a Plaid Guaranteed Payments sandbox. We run six gate checks on a real agent workflow, like a treasury rebalance, P2P payout, or exchange withdrawal. By the end of the week, your team has a working demo and a signed receipt chain anyone can check.
For payment networks looking at agent-originator policies: a HAHS receipt-checking layer in front of one execution lane. The agent sends the transfer with a receipt attached. The network checks the receipt's signature before starting the ACH transfer. The receipt sticks around for the customer's audit trail, the regulator, and the network's own compliance team.