Skip to content

Security, Privacy & AI Governance · CoreSR Inc.

Verifiable controls for evidence synthesis.

Hospitals, universities, HTA agencies, and guideline societies should not need a three-week email round-trip to find out how a vendor protects unpublished reviews, where data resides, or whether AI features send manuscripts to commercial chatbot APIs. Every commitment on this page maps to an active technical control in production.

1 · Certification status

Stated plainly, first: where certification stands today.

CoreSR holds no external third-party ISO 27001 or SOC 2 certificate today. Instead of waiting for an audit firm to tell us to build controls, we built and verified the controls first under an internal Information Security Management System (ISMS) modelled on ISO/IEC 27001:2022.

Today — ISO/IEC 27001:2022-modelled ISMS

Signed security, privacy, and continuity policies, a complete ISO/IEC 27001:2022 Statement of Applicability reconciled continuously against live production controls, a formal Risk Register, quarterly privileged-access reviews, and continuous automated posture verification.

SOC 2 Type II — Observation window Q1 2027

We are pursuing SOC 2 Type II directly rather than paying for a single-day Type I snapshot that institutional reviewers rightly discount. Our Type II observation window is scheduled to open in Q1 2027 (attestation report targeted Q4 2027).

Independent penetration testing

Blocking CI static analysis (SAST), software composition analysis (SCA), secret detection, and recurring dynamic application security testing (DAST) run continuously today. An independent accredited third-party penetration test is commissioned upon institutional procurement requirement or at SOC 2 kickoff (Q1 2027), whichever comes first.

2 · What data CoreSR holds

Published literature and review work product — never patient records.

CoreSR is a research methodology workbench. Knowing what is never in our database is the fastest part of a hospital or guideline privacy review.

Published literature & synthesis work product

Search strategies, bibliographic citations, DOIs/PMIDs, uploaded PDFs of published scientific studies, PICO eligibility criteria, screening decisions, structured extractions, Risk-of-Bias assessments, meta-analytic study tables, GRADE evidence profiles, and manuscript drafts. Customers retain 100% ownership of all uploaded content and generated outputs.

No PHI · No institutional patient records

CoreSR never connects to hospital EMRs, clinical registries, or patient portals. Uploading identifiable patient health information (PHI) regulated under Ontario PHIPA, US HIPAA, or GDPR Article 9 is strictly prohibited under Terms of Service §6.1 and DPA §2.3.

Account fields — and nothing more

Email address, display name, salted adaptive password hash, optional AES-256-GCM encrypted TOTP 2FA seed, optional affiliation, SHA-256 hashed session identifiers, security audit records, and Stripe billing references. Payment card numbers never touch CoreSR servers, and we run zero advertising or cross-site tracking cookies.

3 · Where it is hosted

European Union primary production, Canadian GPU & DR, and US option.

Infrastructure is jurisdictionally clear, encrypted end-to-end, and operated under strict least-privilege governance.

Helsinki, Finland (EU / EEA) — Primary production

Application servers, relational databases, and customer PDF storage run at Hetzner Online GmbH in Helsinki, Finland (ISO/IEC 27001 certified European Union data centres) on dedicated AES-256 encrypted storage behind Cloudflare TLS 1.3 and mutual-TLS origin authentication.

Ontario, Canada — Private GPU & encrypted backup

Primary GPU inference (TITAN-SR, GLASS, ROB BERT) and verified off-site backups run over an authenticated encrypted tunnel on CoreSR-operated infrastructure in Ontario, Canada (PIPEDA jurisdiction; EU Adequacy Decision 2002/2/EC, reaffirmed 2024), with pre-encrypted backup ciphertext replicated to geo-redundant cloud storage and Compliance-mode WORM Object Lock.

US-region option (Team & Institution)

For US healthcare networks, federal/state contractors, or academic centres whose governance mandates US-only residency, CoreSR offers a dedicated United States regional deployment option at the Team and Institution tiers under an Institutional Order Form ([email protected]).

Least-Privilege Administrative Governance: All cloud, source-control, and production administrative access is restricted to individually attributed accounts with 100% hardware/TOTP multi-factor authentication, public-key-only infrastructure access, zero shared credentials, and quarterly access reviews. Institutional reviewers may inspect the Privileged Access Register under NDA via [email protected].

4 · The four control areas

Encryption, network security, vulnerability management, and DLP.

Stated below in the exact words of our downloadable Security Controls Statement (PDF) and DPA Annex II.

1. Encryption

  • In transit: All public HTTPS traffic is terminated at the edge under TLS 1.3 (minimum TLS 1.2 with modern AEAD cipher suites only), HSTS preload, and mutual TLS (mTLS) origin authentication between the CDN edge and CoreSR origin servers. Private east–west traffic between European Union primary production and Canadian GPU and disaster-recovery infrastructure traverses an authenticated point-to-point encrypted tunnel.
  • At rest on production storage: Production relational databases, uploaded study PDFs, extracted review work product, and staging archives reside on AES-256 encrypted storage volumes in ISO/IEC 27001 certified European Union data centres, with cryptographic keys isolated from the operating-system boot volume.
  • Application credentials & secrets: Account passwords are stored exclusively as salted adaptive cryptographic hashes; TOTP multi-factor authentication seeds are encrypted at rest with AES-256-GCM using keys separated from the database; browser session tokens are persisted server-side solely as irreversible SHA-256 digests; optional customer bibliographic API keys are encrypted at rest and masked across all API responses.
  • Off-site backups: Every database snapshot and file archive is asymmetrically encrypted prior to off-site replication using modern authenticated cryptography. Cloud storage subprocessors receive ciphertext only and never hold decryption keys.

2. Network Security

  • Origin isolation & edge protection: Public ingress routes exclusively through global Anycast WAF and DDoS mitigation. Host-level ingress policies default-deny inbound connections, permit HTTPS solely from authenticated edge origins verified by mutual TLS client certificates, and drop direct-to-origin probes.
  • Internal tier isolation: Relational database and internal application workers bind strictly to non-public internal interfaces and are never exposed to the public internet.
  • Hardened administrative access: Infrastructure administration requires individual attributed accounts, cryptographic public-key authentication (password sign-in disabled), brute-force intrusion blocking, and 100% multi-factor authentication across all cloud, source-control, and administrative consoles, reviewed quarterly.
  • Outbound SSRF & HTTP header defences: Server-side URL fetchers validate protocols and resolved network ranges to block internal, private, link-local, and cloud-metadata destinations. Application responses enforce strict Content-Security-Policy, X-Frame-Options, X-Content-Type-Options, Referrer-Policy, and Permissions-Policy headers.

3. Vulnerability Management

  • Continuous CI/CD security gates: Every pull request, main-branch release, and scheduled weekly audit runs blocking static application security testing (SAST), software composition analysis (SCA) across application and ML dependencies, and automated repository secret detection.
  • Automated operating-system patching: Production hosts run enterprise long-term-support Linux distributions with extended security maintenance, automated daily security errata patching, and immediate operator alerting on update anomalies.
  • Recurring dynamic scanning & vulnerability disclosure: Automated dynamic application security testing (DAST) scans run weekly against production, and external security researchers may report findings under safe-harbor terms via https://coresr.ai/.well-known/security.txt ([email protected]).
  • Binding remediation SLAs & penetration testing: Validated vulnerabilities are remediated under written SLAs: Critical ≤ 7 calendar days, High ≤ 30 calendar days, and Medium ≤ 90 calendar days. Independent third-party penetration testing is commissioned upon institutional procurement requirement or at SOC 2 Type II kickoff (scheduled Q1 2027), whichever occurs first.

4. Data Loss Prevention

  • Published-literature boundary (No PHI): CoreSR is engineered strictly for published scientific literature and systematic-review work product; uploading identifiable patient health records or Protected Health Information (PHIPA / HIPAA / GDPR Art. 9) is contractually prohibited and excluded by design.
  • Third-party commercial LLM egress blocked (Team & Institution): For all Team and Institution workspaces, server-side policy blocks outbound calls to external commercial LLM APIs in code and permanently excludes workspace content from model training, confining inference to CoreSR-operated infrastructure (and, only when those private servers are down, stateless zero-retention GPU containers running CoreSR's own models). One exception: when a reviewer runs KM Cor on a Kaplan–Meier figure, the cropped figure, its caption and the review's subgroup names are sent to Google's Gemini vision model.
  • Tenant isolation, MFA & session governance: Every non-public API route enforces authenticated session validation, per-project and workspace role-based access checks backed by automated cross-tenant isolation tests, CSRF protection, and adaptive rate limiting. Team and Institution administrators can mandate workspace-wide TOTP 2FA, enforce shortened session lifetimes, and revoke active member sessions immediately.
  • Audit trail, portability, erasure & immutable resilience: Security-sensitive events write to an append-only audit trail with real-time anomaly alerting. Customers have self-service One-Click Full Account ZIP Export, revocable account deletion followed by automated live purge, and geo-redundant encrypted backups including Compliance-mode WORM Object Lock immutability (RPO ≤ 24 hours, RTO ≤ 24 hours).

5 · AI and your data

Our own models, commercial LLM egress blocked, and pinned provenance.

CoreSR is not a prompt wrapper over OpenAI, Anthropic, or Gemini. We train and host specialist biomedical models — TITAN-SR, GLASS, and ROB BERT — giving institutional workspaces architectural guarantees a wrapper cannot.

Third-party commercial LLM egress blocked (Team & Institution)

Every Team and Institution workspace runs inside a strict server-enforced zero-vendor-egress boundary. Any attempt to invoke an external commercial LLM vendor (OpenAI, Anthropic, Google Gemini, or DeepSeek) fails closed in server code before a network connection can open, with one stated exception: when a reviewer runs KM Cor on a Kaplan–Meier figure, the cropped figure, its caption and the review's subgroup names are sent to Google's Gemini vision model. All other inference runs solely on CoreSR-operated infrastructure; only when those private servers are down does it fail over to stateless, zero-retention GPU containers (RunPod) executing CoreSR's own GLASS v3 weights.

Unambiguous training commitment

Team and Institution workspaces are never used to train, fine-tune, distil, or evaluate CoreSR models (enforced in server export pipelines and warranted in DPA §3.1 and Terms §8.1), unless an institutional research partner executes a separate, bilaterally signed written co-development agreement scoped strictly to its own designated corpus. Individual and Free accounts can opt out in one click under Account Settings → Privacy.

Training data provenance & zero prompt logging

TITAN-SR (v6-2026-10-15) learned from 3,097,955 public scholarly bibliographic records across 19,926 reviews on Apache-2.0/MIT encoder backbones. GLASS (v3) (6 specialist routes, 65,536-token context) is built on permissively licensed Apache-2.0 open weights, open-access biomedical literature, and verified appraisal worksheets. ROB BERT (v6.1) trained on 4,237 deduplicated RoB 2 worksheets. Operational telemetry records token totals, latency, and model identifiers only — zero prompts, zero responses, zero study text.

Model version pinning & human-in-the-loop control

Every AI output pins its exact model release and cryptographic manifest digest in project records and CSV/ZIP exports, backed by our dated Model Version Changelog (PDF). AI proposals never overwrite human adjudications: candidate AI classifications are stored separately from reviewer sign-off, extractions require locatable verbatim PDF quotes before human verification, and quantitative GRADE domains (Risk of Bias, Imprecision, Publication Bias) are computed 100% deterministically in code.

6 · Subprocessors

Every subprocessor, what it touches, and 30-day advance notice.

CoreSR engages six infrastructure and operational subprocessors across all tiers. RunPod runs only CoreSR's own models, and only when CoreSR's private servers are down — see the full Authorized Subprocessor Register (PDF).

CoreSR Inc. Authorized Infrastructure and Operational Subprocessors
SubprocessorRole & Processing ActivityData ScopeRegionCertifications & DPA
Hetzner Online GmbHPrimary compute, relational database & AES-256 encrypted storageAccount metadata and systematic-review project contentHelsinki, Finland (EU / EEA)ISO/IEC 27001 · GDPR Art. 28 DPA
Cloudflare, Inc.Authoritative DNS, Anycast CDN, WAF/DDoS mitigation, TLS 1.3 & mTLS origin authenticationHTTP request metadata and transit-encrypted payloads; static asset caching onlyGlobal Anycast EdgeSOC 2 Type II · ISO 27001 / 27701 · DPA
RunPod, Inc.Standby only: stateless GPU containers running CoreSR's own GLASS weights, used only when CoreSR's private servers are down or unavailableVolatile in-memory inference context only — zero persistent disk, zero prompt logging, zero model trainingUnited States / European Union GPU regionsSOC 2 Type II · Privacy & Data Terms
Backblaze, Inc.Immutable disaster-recovery object storage (Compliance-mode WORM Object Lock, write-only credentials)Pre-encrypted backup ciphertext only — zero access to decryption keys or plaintextUS / Canada / EU regionsSOC 2 Type II · Backblaze DPA + SCCs
Google LLC (Workspace & Cloud)Secondary pre-encrypted backup replica, transactional email delivery & optional Google OAuth 2.0 sign-inPre-encrypted backup ciphertext; transactional email headers/notices; optional OAuth email identifierGlobal (US / Canada / EU)SOC 2 Type II · ISO 27001 / 27017 / 27018 / 27701
Stripe, Inc.Subscription billing, invoicing, tax calculation & payment card processingBilling contact details, subscription tier & card data (entered directly on Stripe; never touches CoreSR)United States & Ireland (EU)PCI-DSS Level 1 · SOC 1 & SOC 2 Type II
30-Day Advance Notification Undertaking: CoreSR provides at least 30 calendar days' advance written notice to Team and Institution administrators before authorizing any new Subprocessor or materially changing processing scope or location, with a contractual 14-day objection and pro-rata refund termination right (DPA §5.3 · [email protected]).

7 · Resilience & continuity

Immutable WORM backups, tested restores, and executive continuity.

Backups only count if compromised credentials cannot delete them and clean-room recovery is routinely verified.

3-2-1-1 encrypted & Compliance-locked backups

Daily database snapshots and study archives are replicated off-host, asymmetrically encrypted before leaving CoreSR infrastructure, stored across geographically separated cloud regions, and locked in Compliance-mode WORM Object Lock using write-only credentials incapable of deleting or truncating retained archives.

Automated restore verification & RPO / RTO ≤ 24h

Automated verification jobs test archive decryption and catalog integrity continuously, backed by automated staleness alerting and quarterly clean-instance disaster-recovery drills. Contractual recovery commitments: RPO ≤ 24 hours · RTO ≤ 24 hours.

Operational continuity & institutional escrow

Security patching, certificate renewal, health watchdogs, and anomaly alerting operate autonomously without manual intervention. CoreSR maintains a formally designated executive Continuity Custodian with escrowed emergency recovery procedures, and Institution plans include an optional Source-Code & Container Escrow Rider.

8 · Incident response & disclosure

72-hour breach notification SLA and safe-harbor disclosure.

Documented playbooks and automated anomaly detection shorten the path from signal to containment.

Incident Response Plan & 72-hour notification

CoreSR maintains a formal Incident Response Plan with dedicated scenario runbooks for Account Compromise, Unauthorised Data Exposure, Credential/Key Compromise, and Ransomware, supported by real-time alerts on authentication anomalies, privilege changes, and bulk export activity.

For any confirmed Personal Data Breach affecting Customer Data, CoreSR notifies affected customers without undue delay and within 72 hours (targeting 24 hours for Institution workspaces) in accordance with GDPR Article 33 and PIPEDA s. 10.1.

Coordinated vulnerability disclosure

Security researchers and institutional assessment teams are authorized under good-faith safe harbor to report vulnerabilities via /.well-known/security.txt (RFC 9116) and [email protected].

Reports are acknowledged within 3 business days, triaged promptly, and remediated within our binding vulnerability SLAs (Critical ≤ 7d, High ≤ 30d, Medium ≤ 90d).

9 · Accessibility

WCAG 2.1 Level A & AA conformance — with a published VPAT 2.5.

Universities, hospitals, and public-sector agencies under Ontario AODA, US Section 508, and EU EN 301 549 require documented accessibility conformance. CoreSR publishes a full ITI VPAT® 2.5 report on CoreSR letterhead directly on this page.

Measured WCAG 2.1 A/AA verification

Automated accessibility audits pass with 0 accessibility violations across 20 public and core workflow views. CoreSR enforces ≥ 4.5:1 text contrast, ≥ 3:1 interactive and focus indicators, skip-to-content navigation, unique per-route titles (WCAG 2.4.2), keyboard-operable screening and extraction workflows, and accessible HTML/CSV data tables accompanying every forest, funnel, and network figure.

Download the VPAT® 2.5 ACR (PDF)

Review our criterion-by-criterion WCAG 2.1 Level A and Level AA Accessibility Conformance Report ready for university and hospital procurement portals:

Download CoreSR VPAT® 2.5 (PDF)

10 · Downloads & NDA pack

Self-service procurement documents (PDF).

Every public legal, privacy, accessibility, and AI governance instrument below is published as a signed CoreSR Inc. letterhead PDF — ready to open in your browser or attach directly to an institutional review ticket with no sign-in gate.

Institutional Vendor Information Pack

Complete executive procurement brief covering corporate identity, hosting residency, security controls, AI governance, subprocessors, RPO/RTO, continuity, and accessibility.

Security Controls Statement

Authoritative specification of Encryption, Network Security, Vulnerability Management, and Data Loss Prevention.

Reusable PIPEDA & GDPR Art. 28 DPA

Pre-drafted Data Processing Addendum incorporating the 2021 EU Standard Contractual Clauses (Module Two) and UK IDTA, formatted on CoreSR letterhead for counter-signature.

VPAT® 2.5 Accessibility Conformance Report

Criterion-by-criterion WCAG 2.1 Level A and Level AA conformance report (AODA / EN 301 549 / Revised Section 508 aligned).

AI Governance & Provenance (Controls 4.1–4.6)

Training data provenance, licensing basis, customer-content training exclusions, zero prompt logging, model pinning, and zero-egress architecture.

Model Version Changelog

Dated production release log, artifact provenance bindings, and PRISMA citation guidance for TITAN-SR, GLASS, and ROB BERT.

Authorized Subprocessor Register

Annex III Subprocessor schedule, processing locations, security certifications, and 30-day advance notification undertaking.

Institutional Questionnaire Response Pack

Standardized procurement Q&A for hospitals, universities, HTA bodies, and clinical guideline societies.

Full ISMS Policy Suite & Detailed Technical Diligence under NDA: Institutional security teams requiring CoreSR's complete internal ISO/IEC 27001:2022 policy suite, Statement of Applicability schedule, Privileged Access Register, Disaster-Recovery Drill Evidence, or detailed technical architecture briefing may request the NDA diligence bundle from [email protected] or [email protected]. See also our Privacy Policy and Terms of Service.