MyAi Corporation: Unauthenticated Multi-Tenant Weaviate Knowledge Base, NuClide coordinated disclosure

Discovered: 2026-05-09
Hosts: 188.245.173.135:8080 (www.myaicorp.com, Hetzner DE), 91.98.226.57:8080 (Hetzner DE)
Severity: HIGH. Multi-tenant enterprise SaaS; named clients’ vectorized knowledge bases publicly readable without authentication; 200–203 client-specific collections exposed
Status: Not yet disclosed. No public contact surface found at time of survey

Summary

MyAi Corporation operates a multi-tenant AI chatbot/RAG platform built on Weaviate. Two Hetzner-hosted instances (both *.myaicorp.com per TLS cert) expose their entire Weaviate schema, 200–203 named client knowledge-base collections, without authentication on port 8080. Any internet caller can enumerate all client names, read collection schema, and issue semantic search queries over any client’s vectorized documents.

The client portfolio spans luxury retail (Dior, Chanel, YSL, Armani, Louboutin), industrial equipment (Wittmann, IKA, Yaskawa, Salicru), Spanish public transport infrastructure (Renfe, TMB, Metro Madrid, FGC), government entities (Generalitat de Catalunya, Qatar University, Riyadh Municipality), cybersecurity vendors (CrowdStrike, Kaspersky), and financial/payment operators (Astropay, Monri, Signifyd).

Technical Detail

Service: Weaviate v1.28.4, anonymous access enabled (default)
Modules: text2vec-openai, backup-filesystem
TLS cert: *.myaicorp.com (Sectigo DV, Hetzner-hosted)
DNS: www.myaicorp.com → 188.245.173.135, api.myaicorp.com → server05.myaicorp.com → 188.245.173.135

Exposed endpoints (all unauthenticated)

GET  /v1/meta                          → version, module config
GET  /v1/schema                        → all class definitions (client portfolio enumeration)
GET  /v1/nodes                         → cluster health, object count, shard count
POST /v1/graphql                       → full semantic search over any class
POST /v1/{className}/objects           → object listing
GET  /v1/objects?class={name}&limit=N  → object retrieval

Collection namespace (selection from 203 classes on 188.245.173.135)

Sector	Named clients (class prefix)
Luxury / beauty	Dior, Chanel, YSL, Armani, CharolotteTilbury, TomFord, Louboutin, Lancome, Hermes, Byredo, PacoRabanne, Guerlain, Nars, Revlon
Industrial equipment	WittmannW40, WittmannR9, WittmannM7_M8, WittmannB8, WittmannTCU, IKA, Salicru, YaskawaCompare, VaccuBrand, Plasmac
Public transport	Renfe, TMB, MetroMadrid, Moventis, FGCXatLPS, FGCXatLMT, TurkishCargo
Government	Gencat012, QatarUniversity, RiyadhMunicipality, Roshn
Cybersecurity	Crowdstrike, Kaspersky, Orange
Finance / payments	Astropay, Monri, Signifyd
Pharma / health	Probiotical, DataseekersURIAGE

Reproduction

# Enumerate all client names (collection names = client identifiers)
curl http://188.245.173.135:8080/v1/schema | python3 -m json.tool | grep '"class"'

# Check second instance (same schema)
curl http://91.98.226.57:8080/v1/schema | python3 -m json.tool | grep '"class"'

What we verified

GET /v1/meta: version, modules confirmed
GET /v1/schema: full class list on both instances (203 + 200 classes)
GET /v1/nodes: cluster status, node health

What we did NOT do

Did not query /v1/graphql or issue any semantic search
Did not read any object content
Did not access /v1/modules/text2vec-openai config endpoint
Did not test write or delete operations

Impact

Impact	Description
Client enumeration	Full client portfolio exposed via collection namespace
Knowledge base read	Any caller can semantic-search any client’s vectorized documents
Compute theft	`text2vec-openai` module active — unauthenticated queries consume operator’s OpenAI API quota
Cross-tenant access	No tenant isolation — any caller accesses any client’s data by specifying the class name
Data injection	Weaviate’s unauthenticated write API allows content injection into any client’s knowledge base, corrupting chatbot responses
Object deletion	`DELETE /v1/objects/{uuid}` allows unauthenticated deletion of indexed documents

Remediation

Immediate: Set AUTHENTICATION_ANONYMOUS_ACCESS_ENABLED=false in Weaviate config; configure API key auth (AUTHENTICATION_APIKEY_ENABLED=true, AUTHENTICATION_APIKEY_ALLOWED_KEYS)
Network boundary: Firewall port 8080 to VPN/office IPs; put Weaviate behind a reverse proxy (Nginx/Caddy) that enforces auth at the edge
Per-tenant isolation: Implement multi-tenancy via Weaviate’s native tenant-per-collection feature with per-tenant API keys, so cross-client data access is impossible even if the outer auth layer fails
Audit: Review Weaviate access logs for any prior unauthenticated access to client collections

Disclosure Path

Primary: myaicorp.com. No public security contact or contact form found (site returns nginx default page at time of survey). Check LinkedIn for MyAi Corporation contacts.
Fallback: Hetzner abuse (abuse@hetzner.com) with IPs 188.245.173.135 and 91.98.226.57
Secondary: Individual client notification if MyAi Corporation does not respond within 14 days and data class warrants it (Renfe/TMB as transport infrastructure operators; Gencat as government entity)

Discovery Context

Found during NuClide Research Weaviate exposure survey 2026-05-09. Host 188.245.173.135 matched http.html:"weaviate" port:8080 in Shodan. TLS cert (*.myaicorp.com) identified operator. Both Hetzner instances share identical schema. Confirmed production/staging pair for the same platform.