Skip to main content

// Security & Trust

Security & Trust at AIErudit

How we protect customer data: the controls we run, the vendors we use, and what is on our roadmap. Written for security reviewers, in plain language.

TLS 1.2+ · HSTSEU app hosting · eu-central-1GDPR-alignedMFA & passkeysPublic status page

// Controls

What we run today.

These statements are intentionally conservative. They describe current platform controls and tracked production-readiness work without implying certifications we do not hold.

Data protection

  • Encrypted in transit with TLS and HSTS on production edge paths
  • Sensitive credentials (MFA secrets, provider keys) are encrypted at rest; managed database and storage layers hold application data
  • MFA and passkeys are available for account protection
  • Card data never touches AIErudit servers; payments are handled by Stripe

Infrastructure

  • AWS eu-central-1 is the primary application and data region
  • Scoped IAM policies and AWS Secrets Manager hold production secrets
  • Database backups and restore drills are tracked as production-readiness work
  • CloudFront and nginx edge controls protect public delivery paths

Monitoring & response

  • Health checks cover app, API, sitemap, and security.txt delivery
  • Better Stack receives error and request metadata for monitoring
  • Public status page at status.aierudit.com is operated outside AIErudit infrastructure
  • Incident-response runbooks and the customer-notification process are tracked as production-readiness work
  • Vulnerability reports follow the public security.txt contact path

Privacy and GDPR

  • Export, delete, and anonymization workflows are built into account operations
  • GDPR Article 33/34 breach-notification runbook defines the 72-hour authority and affected-user notification path
  • For B2B customers, AIErudit commits to notify the customer without undue delay, and no later than 72 hours after confirming a breach affecting customer data
  • Retention and subprocessor commitments are documented for B2B review
  • Analytics uses Mixpanel EU where enabled by consent and configuration
  • DPA review is available for Business and Enterprise procurement

// Subprocessors

Who touches the data.

Every third party that processes personal data, including optional consent-based analytics, session replay and advertising measurement, what they do for us, and where it lives.

Amazon Web Services

EU primary: eu-central-1
Purpose
Hosting, database, storage, CDN origin infrastructure
Data
Application, account, course, and operational data
Region
EU primary: eu-central-1
Transfer
EEA region (eu-central-1); AWS DPA + SCCs cover any provider-side access

Stripe

EU/US
Purpose
Checkout, billing, invoices, refunds
Data
Billing metadata and payment identifiers; card data stays with Stripe
Region
EU/US
Transfer
SCCs / adequacy where applicable

Brevo

EU
Purpose
Transactional and lifecycle email
Data
Email address, name, message metadata
Region
EU
Transfer
EEA processing (Brevo DPA)

Mixpanel EU

EU
Purpose
Consent-aware product analytics
Data
Usage events, country or coarse region, device metadata
Region
EU
Transfer
EU data-residency add-on

Google Analytics 4

Google processing regions (incl. US)
Purpose
Web traffic attribution; loaded only after the Analytics consent category is permitted
Data
Session statistics, approximate location, referrer, browser and device information, interaction events, first-party client identifier
Region
Google processing regions (incl. US)
Transfer
Google data-processing terms + SCCs; advertising storage and personalization stay denied

Microsoft Clarity

Microsoft processing regions (incl. US)
Purpose
Behavioral metrics, heatmaps and session replay; loaded only after the Analytics consent category is permitted
Data
Interaction signals (clicks, scrolling, pointer movement), page-rendering and performance diagnostics; inputs masked, no Identify API
Region
Microsoft processing regions (incl. US)
Transfer
Microsoft data-processing terms + SCCs; session replay retained 30 days

Umami (self-hosted)

First-party host analytics.pagefabrica.com (hosting region to be confirmed)
Purpose
Anonymous public-page traffic, acquisition and funnel analytics
Data
Public path without query or fragment, coarse event properties, transient country and anonymous session hash; no IP stored, no tracking cookies
Region
First-party host analytics.pagefabrica.com (hosting region to be confirmed)
Transfer
No external processor: first-party host under AIErudit data-processing controls

Google reCAPTCHA

Google processing regions (incl. US)
Purpose
Abuse and fraud prevention on protected forms; verified server-side, never used for advertising
Data
Short-lived challenge tokens, request and browser signals
Region
Google processing regions (incl. US)
Transfer
Google data-processing terms + SCCs

Ad platforms: Google Ads, Microsoft Advertising, Meta, X Ads

Provider regions (incl. US)
Purpose
Optional ad measurement and conversion feedback; marketing-consent only and behind product flags
Data
Page-view and standard-event signals, conversion metadata, ad click identifiers; hashed email only where a provider requires matching
Region
Provider regions (incl. US)
Transfer
Provider SCCs / marketing consent; Do Not Sell or Share opt-out honored

Better Stack

EU where available
Purpose
Error monitoring and operational alerts
Data
Error, request, and service metadata
Region
EU where available
Transfer
SCCs where processed outside the EEA

MaxMind

US
Purpose
Server-side GeoIP resolution for consent and localization behavior
Data
IP-derived country or region metadata
Region
US
Transfer
SCCs (DPA in progress)

Google, GitHub, Apple, Facebook

Provider regions
Purpose
Optional OAuth sign-in providers
Data
Provider account identifiers and authentication metadata
Region
Provider regions
Transfer
Provider SCCs / user-initiated sign-in

Anthropic, OpenAI, Gemini

Provider regions
Purpose
Optional AI sandbox and assistant model calls
Data
User-submitted prompts and selected context for active AI features
Region
Provider regions
Transfer
Provider SCCs / data-processing terms

CloudFront CDN

Global edge, EU application origin
Purpose
Edge content delivery and request routing
Data
Request metadata and public asset delivery logs
Region
Global edge, EU application origin
Transfer
Global edge; EEA application origin
Last updated November 7, 202630-day advance notice and objection window for material changes

Subprocessor notices

Subscribe for 30-day advance notices of material subprocessor changes. The form opens as soon as this page finishes loading; until then, email us and we will add you manually.

security@aierudit.com

// Security changelog

Dated trust updates.

Public security and subprocessor changes are listed here, and procurement contacts can subscribe to material subprocessor change notices from the subprocessor section.

security

Password-reset timing, sign-out data clearing, and dependency version floors

The password-reset request now does the same work before it responds whether or not an account exists, so its response time no longer hints at which email addresses are registered. Signing out now clears remaining per-user browser data, such as the saved AI provider key hint, so the next person on a shared device does not see it. Separately, minimum allowed versions were raised for several backend dependencies to prevent resolution of versions with published advisories; the shipped production images already resolved fixed versions. No action is required from customers.

security

JSON Web Token library replaced and organization sign-in token checks tightened

We replaced a third-party JSON Web Token library after a published advisory for which no upstream fix was available. Our review found that our configuration was not affected. We also restricted organization sign-in (OIDC) to an explicit list of signing algorithms and key types. No action is required from customers.

security

SAML federation and enforced single sign-on for Business and Enterprise workspaces are enabled on request

Corrected the identity table: SAML federation and enforced single sign-on are built for Business and Enterprise workspaces, but they are not switched on by default. They are enabled on request, so they now show "On request" instead of "In place".

security

SAML federation built for Business and Enterprise workspaces

SAML 2.0 federated authentication shipped as an identity capability, enabled per workspace on request. Teams can configure IdP metadata, and the ACS endpoint processes signed assertions with real-xmlsec crypto validation, replay protection, and audience restriction enforcement.

security

Enforced single sign-on built for Business and Enterprise workspaces

Team admins can require all workspace members to authenticate through the configured identity provider once it is enabled for their workspace on request. Break-glass recovery and last-admin protection ensure no permanent lockout. Existing sessions receive a grace period before enforcement applies.

security

SCIM 2.0 provisioning available for Business and Enterprise workspaces

SCIM 2.0 automated user and group provisioning is now a supported identity capability. The /api/v1/scim/v2/ endpoint family implements ServiceProviderConfig, Users, and Groups with bearer-token authentication.

security

Customer breach-notification SLA published

Published the accepted B2B customer/controller breach-notification commitment: notice without undue delay and no later than 72 hours after confirming a breach affecting customer data. Kept separate from the GDPR Article 33 regulator clock; DPA template incident clause updated to match.

security

Trust Center Lite public baseline

Published the public security overview, DPA template, security.txt path, certificate verification guidance at /verify/:certificateId, and evidence-backed GDPR erasure posture.

subprocessor

Subprocessor list dated and transfer mechanisms added

Added public subprocessor purpose, data, region, transfer mechanism, and AI-provider data-handling fields.

document

DPA review template published

Added the downloadable Data Processing Addendum template for Business and Enterprise procurement review.

// Enterprise identity

How teams sign in.

Organization identity controls, each described only where implementation evidence exists. In place means switched on today for the plans named on the control (Business and Enterprise workspaces with an active subscription); On request means built, but switched on for a workspace only when you ask.

Organization sign-in (OIDC)

In place

Issuer-configured OpenID Connect sign-in for Business and Enterprise workspaces, with server-side encrypted connection secrets.

Domain verification

In place

Verify a DNS domain to govern workspace membership and controlled auto-join.

Administrative MFA

In place

Administrative access requires TOTP-based multi-factor authentication; passkeys are available for accounts.

Session & device revocation

In place

Members and admins can view and revoke active sessions and trusted devices.

Audit logging

In place

Administrative and user-action audit events are recorded with retention classes.

Microsoft (Entra) sign-in

On request

Microsoft work and school account sign-in for Business and Enterprise users, enabled on request: it is not switched on by default, so contact us to turn it on for your workspace.

Google One Tap sign-in

On request

Ambient Google One Tap sign-in prompt, not switched on by default; it can be enabled per environment and switched off instantly as an incident kill-switch.

SAML federation

On request

SAML 2.0 federated authentication for Business and Enterprise workspaces, enabled on request: it is not switched on by default, so contact us to turn it on for your workspace. Once enabled, teams configure IdP metadata and the ACS endpoint validates signed assertions with real-xmlsec crypto, replay protection, and audience restriction enforcement.

SCIM 2.0 provisioning

In place

Automated user and group provisioning via the SCIM 2.0 protocol, available for Business and Enterprise workspaces. Supports ServiceProviderConfig discovery, user create/replace/patch/deactivate/delete, and group membership sync with bearer-token authentication.

Enforced single sign-on

On request

Team admins can require all workspace members to authenticate through the configured identity provider, for Business and Enterprise workspaces, enabled on request: it is not switched on by default, so contact us to turn it on for your workspace. Break-glass recovery codes and last-admin protection ensure no permanent lockout. Existing sessions receive a grace period before enforcement applies.

// AI data handling

What AI sees, and what it doesn't.

Which AI features send data to which providers, and the boundaries we can prove.

Model training

AIErudit does not train any AIErudit-owned model on customer workspace or course content.

Provider retention

Prompts and selected context are sent to the AI provider you use; retention follows that provider’s data-processing terms. We do not assert provider-side retention or training guarantees we cannot contractually back.

AI transparency

AI-assisted features are clearly labeled in-product. Outputs are AI-generated and may be inaccurate (EU AI Act Art. 50 transparency).

Provider keys

Personal AI provider keys are stored server-side encrypted; the browser holds only a non-secret status flag, never the key.

AI subprocessors

Anthropic, OpenAI, Gemini

Features
AI study assistant, sandbox/terminal, and blueprint generation
Data categories
User-submitted prompts and selected course/workspace context
Retention
Per provider data-processing terms; see each provider policy
AIErudit-owned model training
None (AIErudit-owned models)

// Compliance & roadmap

Where we stand - honestly.

No badge wall. This is the current status of each framework, including the ones we have not reached yet.

GDPR

Aligned

Privacy policy, account data rights, subprocessors, 90-day authenticated IP-log retention, 365-day user/admin audit-log retention, GDPR breach-response runbooks, and DPA review are tracked as public trust artifacts.

PCI DSS

SAQ-A (outsourced)

Stripe-hosted SAQ-A scope: card data is collected and processed entirely by Stripe, a PCI DSS Level 1 service provider. AIErudit never receives, transmits, or stores PAN cardholder data.

EU data residency

Primary region: EU

Primary application systems and data storage run in AWS eu-central-1. Some subprocessors (payments, AI, OAuth, and CDN edge) process data outside the EEA under transfer safeguards — see the subprocessor list.

Security testing

Summary on request

A public-safe security-testing and vulnerability-management summary is available through the security review packet. A recurring third-party penetration-test cadence is not claimed until founder-approved.

SOC 2 Type II

In progress

Controls are being formalized against the Trust Services Criteria. No SOC 2 certification is claimed on this page.

ISO 27001

On roadmap

Formal ISO 27001 certification is under evaluation and sequenced after the SOC 2 control program.

// Accessibility

Built to be usable.

We target WCAG 2.1 AA for learning surfaces. We describe our current accessibility posture honestly and do not claim conformance we have not independently verified.

  • Keyboard operable with visible focus states on interactive controls
  • Minimum 44px touch targets on interactive elements
  • Reduced-motion support via prefers-reduced-motion
  • Color-contrast tokens validated for AA-aligned contrast
  • Semantic landmarks, headings, and labels for assistive technology

Target

WCAG 2.1 AA

A VPAT/ACR is on our roadmap; current accessibility status is available on request.

Read the accessibility statement

// Dependency vulnerability management

How we handle dependency advisories.

Dependency vulnerabilities are detected, triaged, and either remediated or formally risk-accepted through automated tooling and audited risk-acceptance registers. Each entry is backed by a dated internal issue and pull request (reference numbers shown).

Dependabot monitoring

Dependabot monitors npm and pip dependencies for new advisories and opens update pull requests on a weekly schedule.

PR SCA gates

Every pull request runs pip-audit and npm audit as blocking CI gates. A PR cannot merge while a known advisory is open unless that advisory is entered in the risk-acceptance register with a justification.

Nightly SCA and SAST

Nightly SAST and SCA lanes run pip-audit and npm audit on the develop branch, catching advisories published after the last release.

Accepted-risk register

Advisories that cannot be remediated immediately (e.g., upstream pins, major-version freezes) are entered into audited risk-acceptance registers with justification, an expiry date, and a tracking issue. These are disclosed publicly as accepted, never as fixed.

Response times: past performance and current target

Our target is to fix or formally risk-accept critical and high-severity advisories within 8 days of detection. This is a target, not a contractual service level, and we have not always met it. Of the fixes that have reached production since June 2026, the median time from detection to production was 4 days. Two advisories missed the target: the critical Pillow advisory (GHSA-4558-3h8r-v8xp) detected on 20 October 2026, merged 29 October 2026, reached production on 2 November 2026 (13 days); and the high-severity starlette advisories (PYSEC-2026-161) detected on 4 August 2026, risk-accepted at the time, fixed on 29 October 2026, and in production on 2 November 2026 (90 days). Risk-accepted advisories (listed below) are not counted in the production figures.

Accepted advisories are reviewed at the documented review date or when upstream publishes a viable fix, whichever comes first.

Last updated: 2026-11-07

// Recent remediations

Remediation log

16 fixed · 1 accepted / deferred

criticalFixed
ref #14398

PyJWT

GHSA-ffc3-869f-jxw9

The lowest allowed version was raised to exclude the vulnerable range. The shipped production image already resolved a fixed version, so the vulnerable version was not running; the floor pin prevents a future regression.

Detected
2026-11-07
In production
Pending release
Fix
>=2.15.1 (floor pinned)
highFixed
ref #14398

dnspython

CVE-2026-114605

The lowest allowed version was raised to exclude the vulnerable range. The shipped production image already resolved a fixed version, so the vulnerable version was not running; the floor pin prevents a future regression.

Detected
2026-11-07
In production
Pending release
Fix
>=2.9.1 (floor pinned)
moderateFixed
ref #14398

joserfc

CVE-2026-119284

The lowest allowed version was raised to exclude the vulnerable range. The shipped production image already resolved a fixed version, so the vulnerable version was not running; the floor pin prevents a future regression.

Detected
2026-11-07
In production
Pending release
Fix
>=1.7.7 (floor pinned)
lowAccepted
ref #13701

deepmerge

GHSA-gcqx-w48m-w2h4

Build-time dependency of the service-worker build tooling only; not shipped in any browser or server runtime artifact and never given user input.

Detected
2026-10-28
Review date
2026-11-27
Fix
accepted

No patched release exists; the only alternative is downgrading the PWA build plugin, which would break the service-worker build. Accepted for a short, dated period and re-reviewed.

criticalFixed
ref #13957

Pillow

GHSA-4558-3h8r-v8xp

Upgrade to Pillow 13.0.0 fixes GHSA-4558-3h8r-v8xp, GHSA-8g5g-8vmr-9782 and the 37 Pillow 12.3.0 advisories previously accepted (39 in total). Image uploads keep the PNG/JPEG/WEBP decoder allowlist, EXIF size cap and MPO demotion as defence in depth.

Detected
2026-10-20
In production
2026-11-02
Days to production
13
Fix
>=13.0.0
criticalFixed
ref #12814

python-jose

CVE-2026-85394

OIDC algorithm allowlist limited the attack surface; library removed entirely as it is unmaintained.

Detected
2026-10-06
In production
2026-10-08
Days to production
2
Fix
removed (replaced with PyJWT)
moderateFixed
ref #12567

dompurify

GHSA-p98j-92pf-mc4p

DOM XSS advisory in sanitization; mitigated by existing afterSanitize hooks and string-only render paths.

Detected
2026-09-30
In production
2026-10-02
Days to production
2
Fix
3.4.16
moderateFixed
ref #12547

brace-expansion

brace-expansion high advisory (ref #12547)

Transitive glob-pattern dependency; not runtime exploitable.

Detected
2026-09-30
In production
2026-10-02
Days to production
2
Fix
patched via lockfile
highFixed
ref #11394

fast-uri

GHSA-5jgf-p345-68v8

Recurrence of fast-uri advisory family; transitive, not directly exploitable.

Detected
2026-09-02
In production
2026-09-09
Days to production
7
Fix
patched via lockfile
moderateFixed
ref #11386

qs

qs moderate advisories (ref #11386)

Transitive query-string parsing dependency; not directly used in application request handling.

Detected
2026-09-02
In production
2026-09-09
Days to production
7
Fix
>=6.16.0 via lockfile override
highFixed
ref #11308

browserslist

GHSA-73wf-gq98-2v4g

Build-time dependency only; not present in the production runtime bundle.

Detected
2026-09-01
In production
2026-09-09
Days to production
8
Fix
>=4.28.7
highFixed
ref #9931

nanoid

GHSA-2v37-7h3g-55p8

Transitive build-time dependency; not runtime exploitable in the application.

Detected
2026-08-07
In production
2026-08-08
Days to production
1
Fix
>=3.3.8
moderateFixed
ref #9931

dompurify

GHSA-55q2-fjhq-7xh7

Sanitization bypass advisory; mitigated by string-mode render and afterSanitize hooks.

Detected
2026-08-07
In production
2026-08-08
Days to production
1
Fix
bumped to safe version
highFixed
ref #13701

starlette

PYSEC-2026-161

Covers PYSEC-2026-161, CVE-2026-48818, CVE-2026-48817, CVE-2026-54283 and CVE-2026-54282. The resolved starlette already met the fixed versions; the backend now pins starlette>=1.3.1 so the accepted entries could be retired.

Detected
2026-08-04
In production
2026-11-02
Days to production
90
Fix
>=1.3.1 (floor pinned)
highFixed
ref #7642

cryptography

CVE-2026-69247

Bleichenbacher oracle in pkcs7_decrypt; AIErudit does not use pkcs7_decrypt in any code path.

Detected
2026-08-04
In production
2026-08-08
Days to production
4
Fix
>=50.0.0
highFixed
ref #9005

fast-uri

GHSA-4c8g-83qw-93j6

Transitive dependency in JSON schema validation; host confusion and SSRF vectors not exploitable in the application context.

Detected
2026-07-21
In production
2026-07-26
Days to production
5
Fix
patched via lockfile overrides
highFixed
ref #3869

undici

GHSA-vmh5-mc38-953g

TLS certificate validation bypass in SOCKS5 ProxyAgent; undici is a dev dependency (jsdom/vitest), not in the production bundle.

Detected
2026-06-18
In production
2026-06-20
Days to production
2
Fix
>=7.28.0

Accepted advisories

The entries marked "accepted" above are advisories that cannot be remediated immediately due to upstream dependency constraints or a release freeze. Each carries a documented rationale, a review date, and a tracking issue. They are disclosed here honestly and are not claimed as fixed. Our pip and npm risk-acceptance registers currently hold 1 accepted advisory in total, each with a justification and a review date.

Found a vulnerability? See our vulnerability disclosure policy.

// Vulnerability disclosure

Found something? Tell us.

We welcome good-faith security research. Report privately, give us reasonable time to fix, and do not access data that is not yours.

  • Report tosecurity@aierudit.com
  • AcknowledgementTarget: within 2 business days
  • Triage & severityTarget: within 5 business days
  • RecognitionNo bounty program yet; public credit is opt-in

In scope

  • aierudit.com and its subdomains
  • The AIErudit web application and public API

Out of scope

  • Denial-of-service or volumetric testing
  • Social engineering of staff, partners, or users
  • Physical attacks against offices or staff
  • Third-party providers (AWS, Stripe, Anthropic, OpenAI, and similar)

Safe harbor

If you make a good-faith effort to comply with this policy, we will consider your in-scope testing authorized and work with you to resolve the issue promptly. We do not intend to initiate legal action for authorized, in-scope research that complies with this policy; we cannot waive or bind the rights of third parties.

// Breach SLA

When something goes wrong, you hear from us fast.

Our B2B customer/controller breach-notification commitment, stated plainly. This is the contractual notice you receive, separate from your own regulator obligations.

Customer notification commitment

For B2B customers, AIErudit will notify the customer without undue delay, and no later than 72 hours after confirming a breach affecting customer data.

This customer/controller notice is separate from the customer’s own GDPR Article 33 supervisory-authority deadline. Where AIErudit acts as a processor, it assists the customer-controller in meeting that authority deadline but does not file the Article 33 notice on the customer’s behalf unless separately agreed.

Commitment

≤ 72 hours

after confirming a breach affecting customer data, without undue delay.

See the DPA incident clause

// Evidence status

Evidence status and glossary.

Requestable evidence, legal-review gates, and roadmap items are separated so buyer-facing claims stay useful without overstating what has shipped.

Closed engineering tickets are implementation evidence. Public legal or certification claims still require the named gate.

Security testing summary

A public-safe testing and vulnerability-management summary can be shared during qualified procurement review.

On request

This covers SCA and current security-review posture. A recurring third-party penetration-test cadence is not claimed until founder approval exists.

Evidence
#4941
Gate
Founder / Security
Request current packet

DPA template access exists, but the fuller SCC and insurance packet stays in legal/founder review until the request path is confirmed.

Legal review

Do not treat this as a finalized agreement, completed SCC annex set, or insurance certificate. Qualified buyers can ask for the current review packet.

Evidence
#4935
Gate
Founder / Legal
Request current packet

Public Trust Center wording stays conservative until founder/legal approval records a stronger claim.

Legal review

Closed engineering issues are implementation evidence, not automatic legal approval. Unsupported certification or compliance wording remains blocked.

Evidence
#4937
Gate
Founder / Legal
Glossary for procurement and security terms
DPA
Data Processing Addendum or Agreement: contractual data-processing terms between a customer/controller and provider/processor.
SCC
Standard Contractual Clauses: EU-approved clauses commonly used as safeguards for restricted international personal-data transfers.
Insurance
Procurement evidence for relevant business coverage, such as cyber liability, technology E&O, or general liability.
Pen-test cadence
The planned frequency, scope, vendor model, retest policy, and release gate for penetration tests.
SCA
Software Composition Analysis: dependency and package vulnerability scanning plus remediation workflow.
SCIM
System for Cross-domain Identity Management: an API/protocol for enterprise user and group provisioning or deprovisioning.
SAML
Security Assertion Markup Language: an enterprise federation protocol for exchanging authentication assertions.
KMS
Key Management Service: a managed control plane for encryption/signing keys, key permissions, and rotation controls.
SIEM
Security Information and Event Management: centralized security log and event collection with alerting.
SOC 2
An AICPA reporting framework for service-organization controls. AIErudit does not claim a SOC 2 report on this page.
Subprocessor
A vendor or service provider that processes customer or personal data on AIErudit's behalf.

// Questions

Send us your questionnaire.

Security reviews move faster with a human on the other end. Write to us directly, or check live uptime any time.

AIEruditMaster AI by doing
Operated from Warsaw, Poland