PromptShield exists to harden other companies\' LLM features under SOC 2 / ISO 42001 review — so our own posture has to clear the same bar. This page is the canonical answer to procurement and CISO questionnaires until our SOC 2 Type II is signed.
Audit window: 2026 Q4
Gap analysis: 2026 Q3
Available pre-signature on request
TLS 1.3 only on every public endpoint (HSTS, no downgrade). Customer-supplied auth tokens, scan results, and report PDFs are encrypted at rest with AES-256-GCM (Cloudflare R2 / managed KMS). No plaintext storage of credentials.
Production access is gated by SSO + WebAuthn / hardware-key MFA. RBAC separates customer scan execution from billing and from infrastructure. All console + database access is logged to an append-only audit trail kept ≥ 12 months.
Raw prompts and endpoint responses are processed in-memory and discarded after report generation — they are never written to a database. Retention windows: 30 days (Free teaser PDF), 90 days (Starter / Team / Business PDFs). Auth tokens stored encrypted, deleted 30 days after contract end. Canonical retention table on /privacy.
Scan executions are isolated per tenant: separate execution sandboxes, scoped credentials, and namespaced storage. No customer can read another customer's scan inputs, outputs, or report artefacts. Test endpoints provided by customers are not shared between scan workers.
PromptShield only scans endpoints supplied by an authenticated, contracted customer. Default deny-list blocks .gov, .mil, and known critical-infrastructure domains. Acceptable Use Policy on /terms — customers warrant they have authority to test the target endpoint.
Initial acknowledgement within 24 hours of report. Customer notification of any confirmed unauthorised data access within 72 hours, in line with Art. 33 DSGVO. Post-incident summary delivered to affected customers within 14 days.
Researchers may report vulnerabilities to security@promptshield.ai. We commit to (1) acknowledging within 48 hours, (2) not pursuing legal action against good-faith research conducted within a documented scope, (3) crediting researchers in the public advisory unless anonymity is requested. Full disclosure policy ships with the public bug-bounty programme (planned 2026 Q3).
PromptShield is purpose-built to produce evidence for SOC 2 Type II and ISO 42001 reviews. Our own SOC 2 Type I audit is targeted for 2026 Q4 (Type II 2027 Q2). Until then, this Security page + /privacy + /subprocessors form our authoritative pre-audit disclosure for procurement reviewers. Latest internal control narrative available under NDA on request.
Every third party with potential access to customer data is listed on /subprocessors with name, region, purpose, and DPA reference. We notify enterprise customers ≥ 30 days before adding a new sub-processor with material data exposure.
Public surface served from Cloudflare's global edge. Persistent storage in EU regions by default; US-residency available on Business plans on request. Cross-border transfers covered by SCCs and the EU-US Data Privacy Framework (Cloudflare adequacy).
Last reviewed: 26 April 2026 · v1