Security & compliance

Trust Center

We handle some of the most sensitive data there is: conversations that may involve a child. This page sets out how we protect it, what we commit to, and where our independent audits currently stand.

Independent audits

In progress

SOC 2 Type II

An independent examination of how we actually operate our security controls over a sustained observation window, covering security, availability, and confidentiality. This engagement is underway. We have not yet received a report, and we do not claim certification until we have.

In progress

ISO/IEC 42001

The international management-system standard for responsible AI, covering governance of how a model is built, evaluated, monitored, and corrected. Relevant to us because Aletheia makes consequential judgments about people. This engagement is underway and not yet certified.

If you are evaluating Polycreek for a partnership or procurement review and need the current status, scope, or auditor details, contact us and we will share where things stand.

How we handle conversation data

Retention the customer controls

A conversation sent to the Aletheia API is scored and then discarded. We do not keep the text. Where a customer enables audit logging, we store a SHA-256 hash of the conversation plus metadata such as score, threshold, and model version, never the content itself. Bristle, our trust and safety platform, keeps text only as its customer chooses: encrypted, for 30 days by default and never more than 90, deleted automatically when that period ends, and deleted the same day if the customer switches retention off.

No standing human access to content

The Aletheia API retains no content, so there is no mechanism for staff to browse customer conversations; its serving logs contain hashes and scores. In Bristle, Polycreek staff can see a customer's content only through a support session that carries a stated reason, expires on its own, and is recorded in the customer's own audit log. In self-hosted deployments, content never leaves the customer environment at all.

Customer data is never training data

Models are trained on our own corpus. Customer traffic is not used for training or fine-tuning. Detection-improvement telemetry, if a customer opts into it, is limited to hashed, non-reversible signals and is off by default.

We assume a minor is involved

Every input is treated as potentially containing a child's data. We do not build profiles of children and we do not link scores to identities. The platform sending content remains the data controller; Aletheia acts as a processor that returns a score.

Platform security

Encrypted in transit

All API and web traffic is served over TLS. Certificates are managed and renewed automatically.

Scoped API credentials

Every customer holds its own key, with per-customer rate limiting and usage accounting. Keys can be rotated or revoked on request.

Monitored availability

Service health is continuously checked and published, with incident and maintenance history on our public status page.

Least-privilege administration

Administrative interfaces are authenticated separately and restricted to the smallest set of people who need them.

Self-hosting available

Customers with strict data-residency or isolation requirements can run Aletheia entirely inside their own environment.

Published uptime

Current and historical service status is public at status.polycreek.org.

Regulatory posture

We design against COPPA, GDPR, KOSA, and the EU Digital Services Act. The Aletheia API's zero-retention design is the main reason this is tractable for it: most obligations that attach to storing children's data do not arise if the data is never stored. Where a Bristle customer chooses retention, the text is encrypted, time-limited to at most 90 days, and deleted on schedule.

Where a customer carries a reporting obligation, Aletheia is built to support fast escalation of high-confidence threats, including a triage path aimed at reporting to NCMEC within 24 hours. The reporting entity remains the platform; we do not file on a customer's behalf by default.

Policy library

Every policy that governs how Polycreek operates. Governance policies are adopted by our Board of Directors. Operational policies cover security, data handling, child safety and AI governance. We list every policy we maintain so you can see what exists, and the complete documents go to partners, platform customers, grantmakers, auditors and donors on request, with security documentation shared under a mutual non-disclosure agreement.

Governance

Security

Need something that is not here?

We will provide the full text of any policy above, plus a signed copy where you need one, our certificate of incorporation, our IRS determination letter, our annual returns, or a completed security questionnaire. Tell us what the document is for and we will turn it around quickly.

Documents and disclosures