Deep report preview

Agent Treasury Controls, 2026

A category report on how agent products should separate runtime spend policy, treasury approval, custody controls, and owner release.

Treasury operationsMarch 24, 20268 live artifactsFull report free on web

Read full report View artifact manifest

Scope

Built for operators deciding which money movements can stay delegated and which ones should reopen finance, custody, or owner review.

Routine bounded spend should stay delegated and policy-enforced instead of constantly reopening finance review.
Wallet top-ups, rebalances, and payouts are authority-extending treasury actions and deserve stronger custody or finance ceremony.
The right split is between runtime policy, custody policy, treasury approval, and fresh owner step-up for scope release.

Methodology

How this report was assembled

Anchored the report in official custody, treasury, workflow, wallet, and step-up documentation from Coinbase, Stripe, Turnkey, Fireblocks, Passage, Cloudflare, and Oracle as of March 24, 2026.

Used one deep research run plus three focused search sweeps to separate routine agent spend from refill, rebalance, payout, and release decisions.

Modeled treasury control as four distinct surfaces: runtime policy, custody policy, finance approval, and fresh owner presence.

Preferred operator decision rules and control ownership over generalized governance language.

Sources

Public evidence used in this preview

Official

Coinbase CDP Wallets overview

Wallet-control surface for agent-facing products and embedded transaction management.

Open source →

Official

Stripe Treasury overview

Treasury primitives and money-movement framing for balances, outbound flows, and financial accounts.

Open source →

Official

Turnkey embedded wallet guide

Operational wallet-control guidance covering delegated, app-controlled, and shared-custody models.

Open source →

Official

Fireblocks custody overview

Institutional custody and policy surface for treasury-heavy workflows and approval boundaries.

Open source →

Official

Passage step-up authentication

Reference for requiring fresh user presence on sensitive release or payout actions.

Open source →

Official

Cloudflare human-in-the-loop

Workflow pause, approval, timeout, and resume patterns for long-running agent systems.

Open source →

Official

Oracle delegate versus reassign

Useful distinction between temporary delegated action and true ownership transfer.

Open source →

Ecosystem

Cerbos authorization in workflows

Application-layer framing for authorization that persists across workflow state transitions.

Open source →

Ecosystem

AI Runtime Security multi-agent controls

Runtime guardrail framing for delegation depth, no-privilege-escalation, and scope inheritance.

Open source →

Key findings

What the free preview already shows

Finding 01

Most agent products blur runtime spend policy, custody controls, and finance approval until a refill or payout exposes the gap.

Finding 02

Routine API or tool spend should usually stay inside runtime caps and allowlists rather than reopening finance review every time.

Finding 03

Top-up, rebalance, and payout flows extend authority or move cash across risk boundaries, so they need stronger human ceremony than routine execution.

Finding 04

Private-route or policy expansion is a governance event and should require diff-aware review plus fresh owner presence.

Dataset summary

Compact report metrics

Deep Research Runs: 1
Normalized Sources: 79
Public Sources: 9
Sample Rows: 5
Search Queries: 3
Window: Q1 2026

Preview excerpt

Public markdown slice

What this report covers

Where routine agent spend should stay delegated and policy-enforced
Why wallet top-ups, rebalances, and payouts should reopen finance or treasury review
How custody, workflow policy, approval, and step-up fit together without collapsing into one vague control

Core takeaway

Treasury control is not one approval checkbox. Serious agent products need a clearer split between quiet runtime spend, authority-extending treasury actions, and outward-facing policy changes. Routine tool or API spend should remain boring inside caps and policy. Funding, bridging, payouts, and scope release should not.

Why this slug matters

The workflow and payment layers are now strong enough to support a dedicated treasury memo instead of burying the topic inside broader approval or wallet reports. This report turns treasury control into an operator decision framework rather than a generic “more governance” slogan.

Sample rows

What the structured payload looks like

Sample 01Routine tool or API spend

Surface: Routine tool or API spend
Dominant Control: Runtime spend caps plus policy checks
Likely Owner: product operator
Why It Matters: Quiet recurring work should stay delegated while the scope, destination, and budget envelope remain valid.

Sample 02Wallet top-up or refill

Surface: Wallet top-up or refill
Dominant Control: Finance approval plus custody policy
Likely Owner: treasury owner
Why It Matters: A refill extends future authority and changes what the agent can keep doing later.

Sample 03Cross-rail rebalance or bridge

Surface: Cross-rail rebalance or bridge
Dominant Control: Custody controls plus treasury review
Likely Owner: treasury operator
Why It Matters: Moving funds across rails changes settlement, counterparties, and failure modes, not just balance location.

Sample 04Customer payout or refund

Surface: Customer payout or refund
Dominant Control: Fresh finance review plus destination checks
Likely Owner: finance owner
Why It Matters: Outbound transfers are harder to reverse and need named accountability.

Sample 05Private-route or policy expansion

Surface: Private-route or policy expansion
Dominant Control: Owner step-up plus diff-aware review
Likely Owner: owner or security lead
Why It Matters: Changing what the system may touch is a release event, not routine runtime spending.

Artifacts

Paid deliverables for this slug

Full markdown reportFree

Human-readable treasury dossier with the full narrative, charts, and operator recommendations.

LiveMARKDOWN

Endpoint: /api/reports/agent-treasury-controls-2026/markdown

Read full report→

Full machine-readable JSONFree

Structured treasury control rows, source mappings, and governance summary metrics.

LiveJSON

Endpoint: /api/reports/agent-treasury-controls-2026/json

Open artifact→

Chart data artifactFree

Structured chart payload backing the inline report visuals and machine-readable consumers.

LiveCHARTS

Endpoint: /api/reports/agent-treasury-controls-2026/charts

Open artifact→

Definition artifactFree

Saved report definition artifact.

LiveDEFINITION

Endpoint: /api/reports/agent-treasury-controls-2026/definition

Open artifact→

Evidence artifactFree

Structured evidence ledger tying claims and chart provenance back to cited sources.

LiveEVIDENCE

Endpoint: /api/reports/agent-treasury-controls-2026/evidence

Open artifact→

Methodology artifactFree

Structured methodology notes, dataset summary, and report timing metadata.

LiveMETHODOLOGY

Endpoint: /api/reports/agent-treasury-controls-2026/methodology

Open artifact→

Sources artifactFree

Structured source ledger with source kinds, labels, notes, and URLs.

LiveSOURCES

Endpoint: /api/reports/agent-treasury-controls-2026/sources

Open artifact→

Combined report bundleFree

Single purchase target returning the markdown report, JSON artifact, and manifest together.

LiveBUNDLE

Endpoint: /api/reports/agent-treasury-controls-2026/bundle

Open artifact→

About

About this slug

Status: Full report free on web
Source mix: 7 official, 2 ecosystem
Method steps: 4
Version count: 1
Updated: March 24, 2026
Definition: 0 sections, 3 query runners, 2 prompt runners, and 0 chart goals

Saved plan changes are reviewed separately from evidence changes. Query entries become live query runners, and research prompts become grounded prompt runs during create and refresh.