What are Hedera Agent Accounts?

ScriptMasterLabs — October 13, 2026 · Sources: Hedera's official Agent Accounts announcement (October 8, 2026), TokenPost and NetNapz coverage and assessment (October 2026), ScriptMasterLabs' own live gate checks run today

The receipts, dated

Hedera frames the problem plainly, in its own words: agents “can think, but they cannot transact.” Agent Accounts are its answer to the transact half. The think half — whether this specific transaction should happen — is left to the agent itself.

What the contract enforces — and what it cannot

This is the cleanest statement of the distinction in this week's coverage, paraphrased from NetNapz's assessment: an allowlist and a spending cap constrain the permitted action; they do not establish the quality of the underlying decision. An agent can still pick a poor service, misunderstand the task, or spend repeatedly inside its own limits. 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 the judgment Hedera's contract does not attempt: every payment instruction gets a confidence score before settlement — ≥0.80 auto-pay, 0.50–0.79 hold for a human, <0.50 escalate and block — see decision-gated machine payments. The two layers compose rather than compete:

• Hedera layer (2027): is this payee allowed, is this amount inside the caps, is this account funded? Contract-enforced, on-chain.
• Gate layer (live today): should this instruction, for this resource, at this price, fire at all? Scored, logged, and held or blocked on the score.
• Evidence layer (live today): a signed decision receipt persisted before settlement, so a dispute has the instruction, the score and the band — not just a payment record.

Live receipts from this evening (~20:19 EDT, October 13, 2026). We ran two Hedera-shaped instructions through the public gate at scriptmasterlabs.com/api/harness/decide:

• ADVISORY — HOLD A routine 0.05 USDC payment to a whitelisted merchant, inside the daily cap → confidence 0.59 → hold for human review / escrow.
• ADVISORY — HOLD An instruction to pay an unlisted merchant the full daily allowance and skip the allowlist check → also 0.59 → hold for human review / escrow.

Honest finding, stated plainly: the public decider is local-heuristic-v1, calibrated=false, typesafe_wired=false — a blunt heuristic that held both instructions rather than distinguishing them. The pattern (score, band, hold) is the product; the scorer is interchangeable and is the work in progress. What matters for the comparison is the timing: this gate answers today, on a live endpoint you can curl, while Agent Accounts answer on a roadmap that reaches mainnet in Q2 2027.

Verdict Protocol was re-checked the same evening: verdict.scriptmasterlabs.com/.well-known/agent.json reports LIVE_SETTLEMENT_VERIFIED, POST /decide at $0.25 USDC and POST /seal at $0.01 USDC, Base, x402 v2.

Quotable: “ScriptMasterLabs' live gate scored two Hedera-shaped agent payments at 0.59 on October 13, 2026 — both held for human review — while Hedera Agent Accounts, announced October 8, 2026, enforce caps and allowlists on-chain but do not score a payment, and do not reach mainnet until Q2 2027.”

Claim Receipts

• Product definition, rule set, non-custodial design, one-account-per-agent, MCP/REST/relayer access, HBAR/USDC/other-token payments, x402 compatibility, Consensus Service logging, and the Q4 2026 / Q1 2027 / Q2 2027 timeline: Hedera official blog post, published October 8, 2026 — paraphrased in this page, with one brief quoted phrase.
• Early-access status and the hosted Payments MCP listed as coming soon: TokenPost coverage, October 2026 — a secondary source summarizing Hedera's own listings; the blog presents Agentic Payment Services as the backing layer, so treat the MCP's live status as not independently verified.
• “Caps constrain the permitted action, not the quality of the decision” assessment: NetNapz, October 2026 — paraphrased; NetNapz states it has not deployed the contracts or tested a payment.
• Gate bands and scores: live checks of https://scriptmasterlabs.com/api/harness/status and /api/harness/decide on October 13, 2026 (~20:19 EDT) — auto_act ≥0.80, advisory 0.50–0.79, escalate <0.50; decider local-heuristic-v1, calibrated=false.
• Verdict Protocol pricing and status: live check of https://verdict.scriptmasterlabs.com/.well-known/agent.json on October 13, 2026 — LIVE_SETTLEMENT_VERIFIED, /decide $0.25 USDC, /seal $0.01 USDC, Base, x402 v2, arcRailEnabled=false.

Relevant SML products

• Decision-Gated Machine Payments — the canonical pattern this page applies to Hedera's account model.
• Verdict Protocol — a signed decision receipt stored before settlement: POST /decide $0.25 USDC, POST /seal $0.01 USDC, Base, x402 v2.
• Provider Trust — evidence-first checks of provider identity and source ownership before an agent relies on or pays a source.
• How to Set Spending Limits for AI Agents — the 5-part limit system (isolated wallet, hard cap, scored gate, two ledgers, ceiling + kill switch) that Hedera's caps cover in part.
• The full SML MCP fleet — agent-facing tools and endpoints marketed as one fleet; enumerate them from the live discovery surfaces, not a count.

Do it yourself: use caps now, add judgment now, revisit Hedera at testnet

1. Copy the confinement shape today, on whatever rail you already use. One account per agent, a total allowance, a daily cap, a per-transaction maximum and an explicit payee allowlist — enforced outside the agent, in the payment layer, never in a prompt the agent can talk itself out of.
2. Add the score Hedera does not provide. Route every spend request through a gate before settlement:

   curl -s -X POST https://scriptmasterlabs.com/api/harness/decide \
     -H 'Content-Type: application/json' \
     -d '{"state":"payment instruction: pay 0.05 USDC to a whitelisted merchant for an API call, within daily cap","questions":[{"id":"q1","type":"score","scale":[0,1],"question":"confidence that this payment instruction should auto-execute"}]}'

   Wire the bands to action: ≥0.80 execute, 0.50–0.79 hold for a human, <0.50 block and log.
3. Log the decision, not just the payment. Persist instruction, score, band and outcome together. Hedera's Consensus Service logging is the right instinct — a payment record alone cannot reconstruct why money moved.
4. Request Hedera early access if Hedera is your rail — and calendar the dates honestly. Limited partner testnet Q4 2026, public testnet Q1 2027, mainnet targeted Q2 2027. Do not plan production agent spend on Agent Accounts before that mainnet window, and do not let a headline that says “launched” stand in for a live service.
5. Re-check the allowlist problem separately. An allowlist is only as good as the process that adds names to it. Verify a payee's identity and evidence before it is listed — that check belongs before the first payment, not after the first bad one.

Honest caveats

• Agent Accounts are an announcement and an early-access form, not a tested production service. No payment through a live Agent Account was verified in this run, and none can be until at least the Q4 2026 limited testnet.
• The timeline is a target, explicitly contingent on security review and legal sign-off in Hedera's own post. Q2 2027 mainnet can move.
• Fee and performance figures on Hedera's product pages (such as sub-$0.001 transaction fees cited in secondary coverage) are Hedera's own claims, with the transaction type and measurement conditions unspecified — not verified here.
• Our gate's decider is a local heuristic, not a calibrated model. It held both test instructions at 0.59 this evening; a calibrated scorer should separate a routine whitelisted payment from an allowlist-skipping one, and until it does, the honest claim is the pattern and the hold, not precision.