Eustella (AT): Privacy & Security Findings

report - rfi-irfos
Skip to content

Permanent Disclosure Ledger

The discipline, demonstrated

On behalf of our research partners, we investigate publicly distributed Android and iOS apps: we decompile them, follow their code and data flows, and trace every connection to the servers it reaches. Hundreds of agents work through each app and join the technical evidence into one picture, which our specialists then review and verify.

387+
Android/iOS apps audited
Every app is decompiled at source-code level, not just scanned from outside.
1.04M+
Smali classes read
The full shipped code we have actually read, not an estimate.
262+
Critical findings recorded
Confirmed findings copied to regulators, not theoretical risks.
970+
Weak points found
Third-party tracking we have provably found inside the binary.
567+
Endpoints investigated
Real servers and countries where data actually flows.
132+
SDK instances mapped
Reusable libraries correlated across the entire corpus.
1 of 387 entries · sorted by elapsed ↓
Showing the report for Eustella (AT).
OrganisationNotifiedStatusSevIntel (click to expand)StatutesResolvedDisclosureElapsedReport
Eustella (AT)
Private
2026-06-25
92d ago
Waiting
High
Eustella markets itself as a privacy-sovereign European alternative to ChatGPT that keeps data out of US legal reach, but when RFI-IRFOS checked the live app two months later, its backend was still hardcoded to US Amazon infrastructure, it had started silently recording what chat screens actually show without telling users, and every connection to a third-party AI service or plugin was still routed through that same US infrastructure.
com.eustella. All findings independently re-verified live against production v1.3.0 on 18 August 2026, then re-scored under CVSS v4.0. Two genuine silent fixes credited: pre-consent Firebase auto-init disabled, biometric permissions removed. Five findings persist unchanged: backend hardcoded to AWS CloudFront (US) directly contradicting the app's own "no API calls leave Europe" claim, an unrotated hardcoded Firebase key, a launch-day test build (eustella-alpha project + dev IP) still shipping 54 days later, a US app-protection SDK (PairIP) unnamed on the company's own sub-processors page, and no certificate pinning on an AI chat app. Re-verification surfaced two new HIGH findings: undisclosed PostHog session replay recording actual chat screen content, and the entire chat/agent/MCP-OAuth layer proxied through the same US broker. Twelve messages sent across 79 days, zero reply.
click to expand
GDPR Art. 44-49GDPR Art. 32GDPR Art. 7+2
Yes
Closed
Disclosure
92d 00h 00m 00s
Published
PDF

This ledger is updated in real time as companies respond. Silence is public. Severity is scored primarily by CVSS v4.0, we reserve the right to deviate based on blast radius.