We decompile publicly distributed Android and iOS apps and trace their data flows to real endpoints in real countries. We document what we observe, distinguish verified findings from interpretation, and preserve the evidence behind every conclusion. No company can buy a better result — every organisation is held to the same standard, whether it pays us or not. This permanent disclosure ledger puts that principle into practice.
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 and documented
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 Perplexity.
OrganisationNotifiedStatusSEVIntel (click to expand)StatutesResolvedDisclosureElapsedReport
Perplexity
PRIVATE
2026-07-02
107d ago
RESOLVED
HIGH
Perplexity's own text admits that anyone who picks up your phone can use its assistant to send messages and read your notifications without ever unlocking the device, because the app holds sweeping permissions over your SMS, contacts, calendar, phone and email-adjacent data. On top of that, when you speak a query out loud it gets routed through four different outside AI companies, and the app has no way of checking whether the person using it is even old enough to. Separately, the company's own support queue has tried to auto-close this disclosure as resolved three times, each time claiming no reply had been received when one demonstrably had, and twice after its own staff wrote back saying the closure message should not have fired.
ai.perplexity.app.android. The app's own strings admit that "anyone with physical access to your phone can use the assistant to send messages... without unlocking your device" - backed by an extensive OS-level assistant permission surface (SMS, Gmail-adjacent access, Contacts, Calendar, Phone, a system-wide NotificationListenerService). Firebase plus a first-party tracker ContentProvider initialise pre-consent with no consent management platform found, alongside a hardcoded API key. Real-time voice queries route through four separate third-party AI vendors (OpenAI, Google Gemini, ElevenLabs, Soniox) via backend-brokered per-vendor keys. No age-assurance mechanism of any kind. Genuinely good: a verified EU Art. 27 representative (VeraSafe, per Perplexity's own current privacy notice, last updated 8 July 2026), per-capability opt-in connector consent rather than a single "Agree" button, and no Chinese or Russian SDK found. Since this finding first went out, an automated support-ticket closure threat has fired three times on the same thread, word for word identical each time, twice after a named Perplexity support agent wrote back confirming in writing that firing it was not appropriate. Neither retraction changed what the automation did next. A written position arrived on 2026-09-29 after a routing detour through the EU representative; none of the three yes-or-no questions was answered. Case closed and published 2026-10-17, 17 days after the 2026-09-30 embargo.
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. · github.com/rfi-irfos/android-security-audit-2026
this is a useless cookie banner. it's just here to look like one * we don't use cookies, so there's nothing to consent to. don't let anyone tell you otherwise.two buttons, one closes this and throws some confetti. the other literally does nothing.