Security and credentials
Our posture, with dates attached.
Two of these are finished, one is in an observation window, and one has not started. A trust page that cannot tell you which is which is not a trust page.
- Last updated
- 14 February 2026
- SOC 2 Type II
- Report expected May 2026
- Last pen test
- November 2025
- Subprocessors
- 4
Where we actually are
SOC 2 Type II observation window closes in April 2026 and the report should be available in May. ISO 27001 is scoped and not yet started. We are telling you this rather than putting two badges in a footer, because a badge with no date on it is a claim about the past.
Penetration testing is annual against the managed cloud and the collector, most recently by an external firm in November 2025. The summary letter is available under NDA and the two medium findings from it were closed in December.
If a vendor page shows you a certification logo without a date and a scope, ask for both. We would rather you did that to us as well.
How the product is built to fail safely
- Detections run in your process. If our cloud is unreachable, your guardrails keep working and spans queue locally.
- The collector is fail-open for tracing and fail-closed for enforcement, and that choice is configurable per project.
- Payloads can be hashed or redacted at the edge before anything leaves your network.
- Every enforcement override is recorded with the operator, the timestamp, and the span it applied to.
Credentials and controls
| Control | Status | Evidence |
|---|---|---|
| SOC 2 Type II | Observation window open, report expected May 2026 | Auditor letter on request |
| ISO 27001 | Scoped, not started | Roadmap item for H2 2026 |
| External penetration test | Completed November 2025, findings closed | Summary letter under NDA |
| Encryption in transit | TLS 1.3 everywhere, no downgrade path | Continuous, external scan |
| Encryption at rest | AES-256, customer-managed keys on Assurance | Cloud provider attestation |
| Access control | SSO with SAML, enforced MFA for staff, least privilege by default | Quarterly access review |
| Backups | Encrypted, daily, 35 day rolling, restore tested quarterly | Restore log on request |
Subprocessors
That is the whole list. We add to it slowly and we notify customers 30 days before anything joins it that would process trace content.
| Vendor | Role | Region | Touches trace content |
|---|---|---|---|
| Amazon Web Services | Hosting and managed Postgres | eu-central-1, us-east-1 | Yes |
| Cloudflare | Edge network and DDoS protection | Global anycast | In transit only |
| Postmark | Transactional email | United States | No |
| Stripe | Billing for paid plans | United States and Ireland | No |
Responsible disclosure
Send findings to security@sentryllm.ai. Include enough detail to reproduce and give us a way to reach you. We acknowledge within one working day, give you an assessment within five, and we will not send a lawyer after anybody acting in good faith.
In scope: the console, the marketing site, the collector, and the SDK. Out of scope: denial of service, social engineering our staff, and reports generated by a scanner with no manual verification attached.
We do not run a paid bounty yet. We do credit researchers publicly if they want it, and we have sent a fair number of very good bottles of wine.
Asking us things
Security questionnaires go to security@sentryllm.ai and we answer them ourselves rather than routing them through sales. Typical turnaround is four working days for a standard questionnaire, longer if yours has nine hundred rows, which some of them do.