# Safe4

> Safe4 is a payment firewall for AI agents: the security layer between an agent deciding to spend and the money actually moving. It evaluates every agent-initiated payment against four checks — policy, purpose, budget and counterparty — and returns allow, deny or hold before execution, not after settlement.

Safe4 exists because autonomous agents can now initiate payments without a person in the loop. Payment rails and agent payment standards arrived quickly; the control layer that decides which agent payments should execute did not. Safe4 is that control layer.

Key distinctions:

- A payment firewall is not fraud detection. Fraud detection is largely statistical and typically runs after settlement. Safe4 is deterministic and runs before execution, returning a decision the payment cannot proceed without.
- A payment firewall is not an allowlist. An allowlist confirms a counterparty is known. It cannot see whether a payment matches the task the agent was given, or whether repeated identical charges are a loop rather than a decision.
- Safe4 is independent of the settlement rail. The same allow/deny/hold decision applies whether a payment settles in stablecoins, over card networks, or by bank transfer.

## Core pages

- [Safe4 home](https://safe4.ai/): What the payment firewall does and who it is for.
- [Blog](https://safe4.ai/blog): Writing on securing autonomous payments.
- [Terms of Service](https://safe4.ai/terms)
- [Privacy Policy](https://safe4.ai/privacy)
- [Data Handling](https://safe4.ai/data-handling): How Safe4 handles data in transit and at rest.

## Articles

- [Why agent payments need a firewall, not a rule list](https://safe4.ai/blog/why-agent-payments-need-a-firewall-not-a-rule-list): Why allowlists and per-transaction caps fail once an agent, rather than a person, initiates payment; how inspecting intent differs from checking a destination.
- [Four checks before the money moves](https://safe4.ai/blog/four-checks-before-the-money-moves): What the policy, purpose, budget and counterparty checks each ask, and why all four must clear before execution.
- [The control layer above autonomous commerce is still open](https://safe4.ai/blog/the-control-layer-above-autonomous-commerce-is-still-open): Why payment rails and authorization mandates are not sufficient on their own, and what belongs in the control layer above them.

## Developers

- [API documentation](https://api.safe4.ai/docs): Safe4 API reference.
- [GitHub](https://github.com/Safe4AI)