Skip to content

Security model

This page explains how BreachSpider keeps each organization's data apart, how people and programs sign in, what is stored and what is refused. Formal terms are on the legal pages, and the Trust Center summarizes security, data handling and subprocessors.

Tenant isolation: enforced by the database

Every table that holds an organization's data is protected by PostgreSQL row-level security. The API connects as a database role that is not a superuser and cannot bypass these policies. Before any query, the API stamps the caller's organization on the database connection (app.current_org_id) and clears it when the connection is returned; if no organization is set, the policies show nothing. A row is visible only to its own organization, or to the managed service provider (MSSP) that the organization is a client of.

So even a bug in a query that forgot a WHERE organization_id = ... clause would return no other organization's rows: the database itself filters them out.

MSSP accounts work in a client's data through a client context that the server checks against the client's parent organization on every request.

Checked on every release. An isolation gate calls every read route of the API as one organization while another organization's rows exist, and fails the release on any leak. Last recorded full run (2026-11-02): 167 routes, 328 calls, 0 leaks, 0 server errors, 39 foreign identifiers refused.

Signing in: people

  • No passwords. You sign in with a link sent to your email. It works once and expires after 60 minutes. Requests are throttled per address in a way that does not reveal whether an account exists.
  • Two-factor sign-in with an authenticator app, and single-use backup codes, can be turned on per member in Settings. When it is on, it is enforced at every sign-in.
  • Sessions are held in a cookie that scripts cannot read (HttpOnly), that is sent only over HTTPS (Secure) and only to BreachSpider for top-level navigation (SameSite=Lax). A session lasts up to 30 days. You can see every signed-in browser and sign any of them out.
  • Recent sign-in is required for sensitive vault actions (upload, download, setup, delete): within the last 30 minutes.

Signing in: programs

Credential Looks like Can do
API key bs_live_ and 64 hex characters Read, or read and write, as chosen when it was made. Partner and trial keys are limited to the device API routes their plan allows
Hosted MCP connection An OAuth grant you approved in the browser Only the routes the hosted tools call, with the access you approved
Demo token bs_demo_... Public example data only, read only
  • Keys are shown once when made. BreachSpider stores only a SHA-256 hash and the first characters for display.
  • Keys can expire after 30, 90 or 365 days, or not at all. A trial key never outlives the trial. An expired or revoked key is refused with 401 on every route, at once.
  • Management stays with people. Creating or revoking keys, webhooks, billing, account changes and vault changes need a signed-in browser session; an API key, even a write key, is refused.
  • Write scope: a read key that tries to change something gets 403 insufficient_scope.
  • Send keys in the Authorization header only, never in a URL. Store them in environment variables.

Rate limits

Traffic Limit
Device API per key (v1 correlate, v2 Windows) 60 requests and 5,000 devices or hosts per minute by default, shared across v1 and v2
Requests without a key (anonymous, demo) Per address at the edge: 60 API requests per minute with a short burst
Signed-in browsers Per session, sized for the web app's screens
Sign-in links and two-factor codes Throttled per address

Over a limit you get 429 with how long to wait. Details: Rate limits and usage.

What is stored, and what is not

Stored for organizations that keep devices in BreachSpider: each device's name (your own label), vendor, product, version, firmware, part number, type, site, area and zone, and your notes (owner, tags, criticality, role, maintenance window); each device's results, change history and recorded actions; members' email addresses, names, roles and sessions (with the browser and address used to sign in); the audit log.

Never stored for devices: IP addresses and serial numbers. A database trigger clears both on every write, whatever path the data came in by.

Refused outright by the API: device fields that contain an IP, MAC or email address, a serial number or a user name (422, and the value is never echoed back), and an asset_id that is an IP, MAC or email address or a host name.

Not stored by the stateless API calls (correlate-cves, check-windows): nothing about the devices or hosts. Only a request log line (time, key, path, status, count, duration), the per-key rate counters and the monthly count of device checks. Request and response bodies are never logged.

Vault files and workspace backups are encrypted in your browser with keys BreachSpider never has. See Integrity and backups.

What the MCP servers refuse

The MCP servers never send, and the hosted server refuses, these values:

  • host names and fully qualified domain names,
  • IP addresses and MAC addresses,
  • serial numbers,
  • email addresses,
  • user names,
  • site names and locations.

The hosted server names the field and the reason, never the value, and sends nothing. The local server drops such fields before any request and lists them under privacy.dropped_identifying_fields. Use a neutral asset id such as PLC-01.

Outbound connections

When BreachSpider itself connects out (webhook deliveries, batch callbacks, an MSSP's logo), it connects only to public addresses: private, loopback, link-local and other reserved ranges are refused, both when you save the address and when it is used.

Transport and headers

All traffic is HTTPS, through Cloudflare, with HTTP Strict Transport Security. The origin accepts web traffic only from the edge's published address ranges.

The assistant

The assistant (Claude, by Anthropic) is off unless your organization turns it on, never decides results, receives only the facts of the record in question, and every call is logged. See The assistant and its limits.

Reporting a vulnerability

See breachspider.com/security.