Skip to main content
AI PublishMulti-site SEO Operations Platform

Security & Trust

Security Center

The platform is designed around "your keys, your sites, your data": credentials are encrypted per domain, connector requests are individually signed, outbound traffic is restricted, and billing is idempotent and auditable.

Security model: minimal trust boundaries

AI Publish controls risk across three layers:

  • Credentials: AI keys, search platform credentials, CMS connection secrets and payment gateway configuration are encrypted at rest. Interfaces, the admin console and ordinary logs never return plaintext.
  • Connections: the connector file deployed on your server only accepts platform-originated, individually signed, replay-protected requests, and offers no deletion or arbitrary execution capability.
  • Outbound traffic: requests to your sites, image sources and third-party APIs are HTTPS-only, with loopback, private and reserved addresses blocked.

Credential encryption: five isolated key domains

All user credentials are encrypted with AES-256-GCM across five isolated key domains, so a compromise of one domain never exposes the others:

  • AI provider keys (used for your own model calls)
  • Search platform credentials (Baidu, Bing)
  • Email credentials
  • Payment gateway credentials
  • Site integration credentials (CMS connection data)

Every domain supports key-ring rotation: new data is encrypted with the current key while previous keys remain available for decryption, so rotation requires no downtime. Production startup enforces key validation — a missing or invalid key blocks startup instead of degrading to plaintext.

AI keys remain yours: you register and pay the provider directly. The platform stores only an encrypted copy to make the calls you authorize, never bills model usage, and never echoes the full key back to the interface.

Connector protocol: signatures, replay protection, source allow-list

The connector file (aiseo-connector.php) deployed in your site root enforces one shared protocol:

  • Source restriction: only requests originating from the platform domain's resolved egress addresses (or administrator-configured fixed egress IPs) are accepted; everything else receives 403.
  • Request signing (AISEO-HMAC-V2): each request carries an HMAC-SHA256 signature covering the timestamp, a random nonce, the request ID, the action and the SHA-256 digest of the body; timestamps more than 5 minutes off are rejected.
  • Replay protection: nonces are single-use; replaying a signed request is rejected.
  • Size limits: request bodies are capped at 2 MiB and each localized image at 10 MiB.

Connector capabilities are deliberately narrow: reading categories, creating or updating articles, handling images and looking up article status. There is no remote category editing, no remote article deletion, no arbitrary file writes and no arbitrary PHP execution. The signing secret is written once into a protected local file during site enrollment and is never echoed by any platform interface.

Outbound guards: HTTPS-only with address validation

Every outbound connection from the platform follows:

  • HTTPS only (port 443), no HTTP fallback, no cross-host redirect following.
  • Loopback, private, link-local and reserved addresses are blocked to reduce SSRF risk.
  • Remote images are validated for DNS-to-connection IP consistency, HTTP status, Content-Type and size; only jpg / png / gif / webp are accepted.

The same rules apply to site analysis crawling: robots, response-body and timeout limits are respected, and non-HTML resources are skipped.

Billing safety: idempotency and success-only charges

The billing pipeline uses idempotency keys to prevent double charges:

  • Publish and generation operations carry the platform article ID as an idempotency key; connectors use it to avoid duplicate articles on timeout retries, and the platform uses it to avoid duplicate charges.
  • Failures, cancellations and system retries of the same result are never billed; only successful outcomes create platform feature charges.
  • Overage billing is off by default once plan quotas are exhausted; you must explicitly enable it and set a monthly cap.
  • Payments run through the official Alipay and Airwallex gateways with signature-verified callbacks; the platform never stores your payment account password.

Audit trail, boundaries and expiry retention

Important operations — task submissions, configuration changes, publish and submission results — are all recorded. Each task freezes a snapshot of the model, template, targets and pricing at submission time, so later configuration changes never affect the explainability of historical tasks.

We are equally explicit about boundaries: model usage is billed by your AI provider; CMS and search platform availability is decided by third parties, while the platform records, retries and alerts; and we never promise search rankings.

After a plan expires, business data becomes read-only: new submissions and automatic publishing stop. Renewing within 30 days preserves your data; after the retention window, data is cleaned up per platform rules.

Security Center

The platform is designed around "your keys, your sites, your data": credentials are encrypted per domain, connector requests are individually signed, outbound traffic is restricted, and billing is idempotent and auditable.