Trust & security

Security is the product, not a page

ProcessTwin automates work with AI agents — which is exactly why it was engineered against the OWASP Top 10, the OWASP LLM Top 10 and multi-tenant SaaS guidance from day one. Here is precisely what that means.

Tenant isolation, twice

Every workspace is isolated at two independent layers: application-level tenant scoping on every query, and PostgreSQL row-level security underneath it. Cross-tenant requests don’t error — they simply see nothing.

  • Tenant id enforced on every query, cache key and storage path
  • Row-level security policies as defense in depth
  • Automated cross-tenant probe tests in CI on every change

Encryption & secrets

Connector credentials are encrypted with AES-256-GCM using keys derived per tenant — a leaked ciphertext from one workspace is cryptographically useless against another. Master keys rotate without data migrations.

  • Per-tenant HKDF-derived data keys
  • TLS in transit, encryption at rest
  • API keys stored as salted hashes, shown once

Governed AI agents

Agent guardrails are enforced in the execution layer, never just in a prompt. Content from connected tools is treated as data, not instructions — and screened for prompt-injection before an agent ever sees it.

  • Tool allowlists, spend caps and blast-radius limits per agent
  • Human approval gates and an instant kill switch
  • Adversarial evaluation suite gates every deployment
  • Optional PII redaction before any AI provider call

Tamper-evident audit

Every privileged action — human or agent — writes to an append-only audit log. Entries are hash-chained, so any alteration of history is detectable, and the chain can be verified from the product.

  • Hash-chained, append-only audit entries
  • Impersonation sessions require a reason and auto-expire
  • Exportable for your SIEM

UK GDPR & data lifecycle

Data protection is built into the product, not a policy PDF. We operate under UK GDPR and the Data Protection Act 2018, with a signed DPA and published subprocessor register for every customer. Retention windows are configurable per workspace, exports are one click, and deletion is real deletion.

  • Export-your-data: complete workspace archive on demand (Art. 20)
  • Deletion with a 30-day grace period, then hard removal including keys
  • Configurable retention for signals, traces and logs
  • EU hosting (Germany) under UK adequacy; dedicated single-tenant instances on Enterprise
  • No solely automated decisions with legal effect — human approval gates by default (Art. 22)
  • Breach notification: ICO within 72 hours where required, customers without undue delay

Disclosure & operations

We run the same platform we sell — with rate limiting on every surface, security headers on every response, replay-proof webhooks, and continuous dependency and secret scanning in CI.

  • security.txt with a monitored disclosure channel
  • Secret scanning, SAST and dependency audits on every commit
  • Non-root, read-only containers in production
Found a vulnerability? We want to hear about it: /.well-known/security.txt describes our disclosure process. Reports are acknowledged within 48 hours.