AI agents perform tasks that cost money. They book servers, purchase API calls, hire other agents. But financial systems require a human on the other end. Every bank, every payment processor, every compliance framework assumes that a person holds the account, signs the transaction, and bears liability. AI agents fit none of these assumptions.
The Three Gaps
The global financial system handles $150 trillion in annual cross-border flows. Every dollar moves through identity verification, spending controls, and sanctions screening. These three pillars were designed for humans. AI agents break all three.
Gap 1: Identity
Know Your Customer (KYC) requires a government-issued ID, a proof of address, and sometimes a live video call. An AI agent has none of these. It has a model version, an operator, a set of permissions, and a transaction history. No passport office issues documents to software.
Some teams work around this limitation by routing agent transactions through a human account. The human signs up, passes KYC, and the agent spends from that account. This creates a legal fiction. The human becomes liable for decisions they never made, approvals they never granted, and transactions they never reviewed.
Gap 2: Control
Credit cards have spending limits. Corporate accounts have approval workflows. Wire transfers require dual authorization above certain thresholds. These controls exist because humans make mistakes, and organizations need guardrails.
Agents need different controls. A human might overspend on a business dinner. An agent might execute 10,000 microtransactions in a single second. The failure modes differ in kind. A buggy agent can drain an account in milliseconds. A compromised agent can route funds to sanctioned entities at machine speed. Human-designed rate limits and approval workflows cannot contain these scenarios.
Gap 3: Compliance
OFAC screening checks a name against a sanctions list. The screener assumes a human counterparty with a name, a nationality, and a date of birth. When Agent-47B sends $50 to Agent-892C for a completed data labeling task, the sanctions screening pipeline has no input to work with. Neither agent has a nationality. Neither appears on any watchlist. The compliance system returns a null result, and the transaction either blocks (false negative risk) or passes without review (regulatory risk).
| Dimension | Human (KYC) | AI Agent (Today) | AI Agent (Oris KYA) |
|---|---|---|---|
| Identity Document | Passport, Driver License | None | Agent Profile + Operator Attestation |
| Verification Method | Document scan + liveness check | None | Cryptographic signing + behavioral analysis |
| Spending Controls | Card limits, approval chains | Inherits human account limits | Per-agent policy engine (amount, frequency, destination) |
| Sanctions Screening | Name + DOB + nationality match | No applicable data fields | Operator entity + wallet address + trust level |
| Liability | Account holder | Falls to human proxy | Operator (developer) with audit trail |
| Transaction Speed | Minutes to days | Milliseconds | Sub-200ms with full compliance |
Why Workarounds Fail
The most common workaround today involves a shared API key tied to a corporate account. The company passes KYC once, and every agent uses the same credentials. This approach collapses individual agent accountability into a single identity. When something goes wrong, there is no way to determine which agent authorized which transaction. Audit trails become useless. Regulators see a single entity making thousands of contradictory decisions per hour.
Another approach uses prepaid wallets with hard caps. The operator loads $100 into a wallet, and the agent spends until the balance reaches zero. This prevents runaway spending, but it offers no granularity. The agent can spend the entire $100 on a single prohibited transaction. The cap controls volume. It does not control behavior.
Three Gaps, Three Systems
Oris addresses each gap with a dedicated subsystem. KYA (Know Your Agent) solves identity by creating a verifiable profile for every agent, tied to a registered operator. The Policy Engine solves control by enforcing per-agent spending rules that evaluate in under 10 milliseconds. The Compliance Bridge solves screening by mapping agent wallets to entity profiles that sanctions databases can process.
Every agent that transacts through Oris carries a verified identity, operates within enforced boundaries, and passes compliance screening before a single dollar moves.
The agent economy will grow whether or not the financial system is ready. Gartner projects that agentic AI will handle 15% of day-to-day work decisions by 2028. These agents will need to spend money. The question is whether they will spend it through ad-hoc workarounds with no accountability, or through purpose-built infrastructure that regulators and enterprises can trust.
Learn how Oris closes the identity, control, and compliance gaps at useoris.xyz
Get started with Oris
Two minutes to set up. Full spending controls from day one.