Trust Center
What BreachSpider does with the information you give it, how the service is built and run, and what is not in place yet. Every statement on this page describes the service as it is configured today. Where a detail is still to be confirmed, it says so.
On this page
- The short version
- Architecture
- Tenant isolation
- Encryption
- Sign-in, sessions and API keys
- Rate limits and abuse controls
- Audit logging
- Backups, vault and integrity seals
- The assistant
- Data sources and citation policy
- Subprocessors
- Retention
- Compliance status, stated plainly
- How BreachSpider supports common frameworks
- Reporting a vulnerability
- Legal documents and versions
The short version
- BreachSpider never connects to your control network. It does not scan, probe or log in to your devices. It works only from the device details you enter, import or send to the API: vendor, product, version and the context you choose to add.
- Each organization sees only its own records. Customer tables carry the organization they belong to and the database enforces that with row-level security on every request.
- Traffic is encrypted in transit and backups are encrypted at rest. The production database itself is not encrypted by BreachSpider; see Encryption.
- Matching is deterministic. Which CVEs apply to which version is decided by version rules from cited sources, not by a language model. The assistant (Claude, by Anthropic) explains and drafts; it does not decide findings.
- No certifications today. BreachSpider has no third-party audit report or certification. See Compliance status.
Architecture
- Hosting. BreachSpider runs on a single virtual server operated by OVHcloud. Its data center is in the United States (US East).
- Edge. All web and API traffic passes through Cloudflare (content delivery, TLS and denial-of-service protection). The server's firewall admits web traffic (ports 80 and 443) only from Cloudflare's published address ranges.
- Internal services. The PostgreSQL database, the API process, the cache and the vault signing service listen on the server's local interface only; none of them is reachable from the internet.
- Database role. The API connects to PostgreSQL with a restricted role that is not a superuser, so row-level security applies to it.
- Response headers. Pages are served with HSTS (one year, including subdomains), a Content Security Policy, frame blocking,
nosniff, a strict referrer policy and a permissions policy that turns off camera, microphone and location for the site.
Tenant isolation
Every customer record belongs to one organization. Isolation is enforced twice:
- In the application. Each request is resolved to the caller's organization from the session or API key, and every query is scoped to it. A record of another organization generally answers "not found", so its existence is not revealed.
- In the database. PostgreSQL row-level security policies on the customer tables compare each row's organization with the organization set on the connection for that request. A connection with no signed-in caller is set to an organization that does not exist, so a query that runs without a caller returns nothing rather than another customer's rows.
Tests that create two throwaway organizations and try to read across them run in the release checks. MSSP accounts can see the client organizations linked to them, and only those.
Encryption
| Where | What is in place |
|---|---|
| Browser to Cloudflare | HTTPS with TLS 1.2 or 1.3; HSTS. |
| Cloudflare to server | HTTPS with TLS 1.2 or 1.3 to the origin certificate. |
| Database backups | Each nightly dump is encrypted with AES-256 on the server as it is written, before it is copied off the server. It never touches disk unencrypted. |
| Server file backups | Encrypted by the backup tool (restic) before upload. |
| Vault files | Encrypted in your browser before upload (Argon2id key derivation, AES-256-GCM). BreachSpider never receives the passphrase, the recovery key, any key, the file or its name. |
| Workspace backup exports | Encrypted in your browser with a passphrase you choose (AES-256-GCM). BreachSpider cannot open or recover them. |
| Production database | Not encrypted by BreachSpider. The database files sit on the hosting provider's storage. Device fields you enter, including IP addresses or serial numbers if you choose to enter them, are stored without field-level encryption. |
Sign-in, sessions and API keys
- No passwords. You sign in with a one-time link sent to your email address. A link is valid for 60 minutes. Sends to one address are limited to one a minute and five an hour.
- Two-factor sign-in. Any user can turn on an authenticator app code (TOTP) in Settings, with ten one-time backup codes.
- Sessions. A session is a random 256-bit token held in an
HttpOnlycookie that is sent over HTTPS only, withSameSite=Lax, for up to 30 days. The server stores only a one-way hash of it. You can list your sessions and end any or all of them in Settings. - Roles. Owner, admin, analyst, engineer and viewer. Viewers cannot change records. Vault uploads, downloads and setup also need a sign-in within the last 30 minutes.
- API keys. Keys start with
bs_live_, are shown once, and are stored only as a one-way hash. A key can be given an expiry of 30, 90 or 365 days, and can be revoked at any time; an expired or revoked key is refused on every route. Keys are created and revoked only from a signed-in session, never with another key. Each key records when and from which address it was last used. - Connected apps. Apps connected through the hosted MCP sign in with OAuth and receive a connection limited to the device, Windows, CVE, catalog and batch endpoints.
Rate limits and abuse controls
- Request limits per address and per API key at the web server, with a stricter limit on sign-in (five requests a minute per address) and on contact and demo forms.
- Per-key usage is metered and reported in response headers.
- The device check endpoints and the MCP server refuse identifying values. Sent device fields that contain an IP or MAC address, an email address, a serial number or a user name are refused with the field named and the value never echoed. The MCP server also refuses host names and site names. Use a neutral asset id such as
PLC-01instead.
Audit logging
Changes to accounts, team members, keys, devices, records and settings are written to an audit log with the organization, the person, the action, the record affected, the time, and the IP address and browser of the request. Calls to the assistant made for account features are recorded with the model used and the scope of what it read. The audit log can be read in Settings and exported, depending on role and plan. Writing an audit record never blocks the action itself, so a failed audit write is logged as an error for us to investigate.
Backups, vault and integrity seals
Service backups
- A consistent database dump every night at 01:00 UTC, encrypted on the server and copied to Backblaze B2 at 01:30 UTC. Two days are kept on the server and about 30 days off the server.
- A whole-server file backup every night (database files excluded, since the dump covers them), encrypted and kept in Backblaze B2 as 7 daily, 4 weekly and 6 monthly snapshots.
- Recovery point: up to 24 hours (the last nightly dump).
Documents and configuration vault
Where an organization uses the vault, files are encrypted in the browser and stored as ciphertext in a private Backblaze B2 bucket under an object name that is never reused. A separate signing service, running as its own operating system user with its own keys and its own append-only ledger, seals each version: it reads the stored bytes itself, measures their digest and length, and signs a manifest that extends a hash chain per organization. The web application cannot read the signing keys and cannot ask the signer to sign a digest it supplies.
- Hybrid post-quantum seals (since November 7, 2026). New seals are signed with both Ed25519 and ML-DSA-65 (FIPS 204), and verification requires both signatures. Seals made before that date carry Ed25519 only and keep verifying; they are shown as classical seals.
- Daily anchor. Once a day the signer signs the chain heads of every organization and writes that anchor to a separate bucket under Object Lock retention (governance mode, 400 days).
- Verification. Anyone in the organization can re-verify a version at any time. Each check reports Verified, Mismatch or Unable to verify. A download is blocked unless the version verifies; a vault admin can override with a written reason, which is logged.
- What a seal shows. A seal shows that the stored file is the one that was sealed and has not changed since. It does not show that the file's contents are correct.
The assistant
The assistant (Claude, by Anthropic) answers questions, explains findings, maps unfamiliar import columns, suggests products for import rows the engine could not resolve, and drafts tickets and report summaries. It never decides whether a CVE applies to a device: those findings come from the matching engine, and every suggestion the assistant makes for an import row goes through the same engine before a person accepts it.
When you use it, BreachSpider sends Anthropic your message, the earlier messages in that conversation, and the account context the answer needs, which can include device names, site names, products, versions and findings. Conversations are saved to your account so you can return to them, and you can delete them. Anthropic processes this data as our subprocessor under its commercial terms.
If a user asks for an answer to be read aloud, the text of that answer is sent to ElevenLabs to produce the audio.
Data sources and citation policy
BreachSpider's findings are built from public and vendor-published records. Naming a source here describes where data comes from; it does not mean that source endorses BreachSpider.
- CVE records from the CVE Program and the National Vulnerability Database (NVD).
- The CISA Known Exploited Vulnerabilities catalog and CISA ICS advisories.
- Vendor security advisories, in CSAF form where the vendor publishes it, and national CERT advisories.
- Microsoft security update data for Windows builds and updates.
- Exploit prediction scores (EPSS, published by FIRST).
- Vendor lifecycle and firmware pages for end of support and latest firmware.
Citation rules the product follows:
- Every finding names the source of its affected version range and links the vendor advisory when one exists.
- Every lifecycle date links the vendor statement it came from. Only vendor statements are used for lifecycle dates.
- Evidence classes are kept apart: a fix the vendor states is never shown as the same thing as a fix BreachSpider derived.
- No result is ever presented as "no risk". A device with no matching CVE, an unresolved product or partial coverage is reported as exactly that.
More detail: Editorial policy.
Subprocessors
The current list, with what each one receives, is on the Subprocessors page: OVHcloud (hosting), Cloudflare (edge network), Backblaze (encrypted backups and vault storage), Anthropic (the assistant), Resend (email), ElevenLabs (read-aloud audio), Stripe (card payments) and Google Analytics (public marketing pages only).
Retention
Saved reports and the evidence in them are kept for 3 years. Device check history, change events, assistant action records and daily progress snapshots are removed after 3 years. When an organization is deleted, its data is removed within 24 hours; backups age out within about 30 days. The full schedule is on the Data retention page.
Compliance status, stated plainly
- BreachSpider holds no third-party certification and no independent audit report. No SOC 2 report and no ISO/IEC 27001 certificate exist for BreachSpider today, and none is in progress.
- No penetration test report by an outside firm is available.
- Security questionnaires are answered on request, from the facts on this page.
How BreachSpider supports common frameworks
Using BreachSpider does not make an organization meet any standard. These are the specific tasks it helps with; your assessor decides what counts as evidence.
| Requirement area | How BreachSpider helps |
|---|---|
| NERC CIP-007 R2 (security patch management: evaluate applicable patches at least once every 35 calendar days) | Lists, per device and version, the vendor-published fixes and advisories that apply, with sources and dates, and keeps a dated history of each check and of the actions you record. |
| NERC CIP-010 R3 (vulnerability assessments) | A device-by-device list of matching CVEs with the affected range and its source, exportable as a report. |
| IEC 62443-2-3 (patch management in industrial automation and control systems) | Tracks which vendor fixes apply to which installed versions and records what was done, by whom and when. |
| Asset inventory practices in most frameworks | A device inventory with vendor, product, version, site and lifecycle status, imported from your own lists. |
Reporting a vulnerability
Report security issues in BreachSpider to [email protected]. The policy, scope and our commitment to good-faith researchers are in the vulnerability disclosure policy; the machine-readable contact is at /.well-known/security.txt.
Legal documents and versions
| Document | Version | Date | What it covers |
|---|---|---|---|
| Trust Center | 2.0 | November 8, 2026 | How BreachSpider is built and run: isolation, encryption, access, backups and integrity seals, data sources, subprocessors and an honest compliance status. |
| Terms of Service | 2.0 | November 8, 2026 | The agreement for using BreachSpider: what the service is and is not, your data, acceptable use, liability, and disputes. |
| Privacy Policy | 2.0 | November 8, 2026 | What personal data BreachSpider collects, why, who receives it, how long it is kept and your rights. |
| Cookie Notice | 1.0 | November 8, 2026 | Every cookie and browser storage item BreachSpider sets, what it is for and how long it lasts. |
| Acceptable Use Policy | 2.0 | November 8, 2026 | What you may not do with BreachSpider, permitted security work, and how we enforce the policy. |
| Data Processing Addendum | 2.0 | November 8, 2026 | How we process customer data for you as a processor: instructions, subprocessors, security measures, incidents, deletion and audits. |
| Subprocessors | 1.0 | November 8, 2026 | The companies that process data for BreachSpider, what each receives and why. |
| API and SDK Terms | 1.0 | November 8, 2026 | Terms for the BreachSpider API, the Python SDK, the MCP server and the widget: keys, limits, versioning, attribution and what you may build. |
| Data Retention Schedule | 1.0 | November 8, 2026 | How long each kind of record is kept, what removes it, and what happens when an organization is deleted. |
| Security and Vulnerability Disclosure Policy | 2.0 | November 8, 2026 | How to report a vulnerability in BreachSpider, what is in scope, what we commit to, and where the security overview lives. |
| Service Level Agreement | 1.x | September 28, 2026 | Not revised in this set; under review by the owner. |
Changes between versions are listed on the Legal page.
