What are Hedera Agent Accounts? Caps on-chain, mainnet in Q2 2027

Hedera Agent Accounts, announced October 8, 2026, are programmable, non-custodial spending accounts for AI agents. A user funds an account for one designated agent from an existing Hedera wallet, sets a total allowance, a daily cap, a per-transaction maximum and an allowlist of merchants and recipients, and those rules are written to a Hedera smart contract. The agent requests spending; the contract validates and executes it. Private keys never leave the user's wallet, funds do not move until a purchase occurs, and every action is logged to Hedera Consensus Service. Access is via Agentic Payment Services — an MCP server, a REST API and a relayer — with payments in HBAR, USDC or other Hedera tokens, and compatibility with x402 purchases.

They are not live on mainnet. Hedera's own timeline is limited partner testnet in Q4 2026, public testnet in Q1 2027, and mainnet targeted for Q2 2027, pending security review and legal sign-off. TokenPost notes the offering is presented through an early-access form, with no general-availability date announced.

What caps enforce — and what they cannot

• A total allowance and daily cap stop an unbounded drain. They do not stop a slow bleed of small, permitted, wrong payments.
• A per-transaction maximum stops one oversized purchase. It does not stop many right-sized purchases of the wrong thing.
• An allowlist stops payments to unknown addresses. It does not stop a bad purchase from a listed merchant — the list proves permission, not quality.
• A Consensus Service log settles disputes about what happened. It does not prevent a bad payment before the fact.

An allowlist and a spending cap constrain the permitted action; they do not establish the quality of the underlying decision. Hedera has built the confinement layer well. Confinement is not judgment.

The missing layer is a score, applied per payment

The decision-gate pattern adds what the contract does not attempt: confidence ≥0.80 auto-pays, 0.50–0.79 holds for a human, below 0.50 escalates and blocks.

On October 13, 2026 (~20:19 EDT) we ran two Hedera-shaped instructions through the live gate at scriptmasterlabs.com/api/harness/decide. A routine 0.05 USDC payment to a whitelisted merchant inside the daily cap scored 0.59 → advisory, hold for human review. An instruction to pay an unlisted merchant the full daily allowance and skip the allowlist check also scored 0.59 → advisory, hold.

Honest finding: the public decider is local-heuristic-v1, calibrated=false — a blunt heuristic that held both rather than distinguishing them. The pattern is the product; the scorer is interchangeable. What matters for the comparison is timing: the gate answers today on a live endpoint, while Agent Accounts answer on a roadmap reaching mainnet in Q2 2027.

Verdict Protocol was re-checked the same evening: LIVE_SETTLEMENT_VERIFIED, POST /decide $0.25 USDC, POST /seal $0.01 USDC, Base, x402 v2.

Do it yourself

1. Copy the confinement shape today on your current rail: one account per agent, total allowance, daily cap, per-transaction maximum, explicit payee allowlist — enforced in the payment layer, never in a prompt.
2. Add the score: route every spend request through a gate before settlement and wire the bands to action.
3. Log the decision, not just the payment: instruction, score, band and outcome together.
4. Request Hedera early access if Hedera is your rail — and calendar Q4 2026 / Q1 2027 / Q2 2027 honestly. Do not plan production spend on a headline that says "launched."

Full analysis with claim receipts: https://scriptmasterlabs.com/hedera-agent-accounts

Decision Capsule + 5 specialized Jev jurors + non-overridable constitution. Jev chooses. Code decides. Built for agents that manage real capital.