how we work.
machines scan. people verify. nothing is reported until a person has confirmed it, and nothing is published while a flaw is live.
four steps, every app.
the same route for every app. the numbers under each station are what reached it across the pilot so far.
- 35,854scanstatic analysis of the public build
- 384verifya person confirms it is real and live
- 384reportprivately, to the owner or the cert
- 19fixre-verified on the next build
- 01
scan
proactive, at scalelarge-scale static analysis of public apks across countries and sectors, including the apps that run no security program at all.
- public builds only, fetched the way any user would
- no login, no account, no interaction with a live service
- about 14.4 apps a minute per worker; a sweep of 1,500 apps takes under two hours
- 02
triage
human verificationevery candidate is confirmed by hand. test keys, sample configs and public identifiers are dismissed before anything is reported.
- is it live? is it really a secret? does it really belong to this owner?
- placeholders, sdk samples and public identifiers are thrown out
- in a recent sweep, 540 flags became 6 reports
- 03
report
responsible disclosurewe write to the owner privately, or to the national cert where the owner is a public body, with evidence and remediation guidance.
- what we observed, where, and how to confirm it
- a redacted copy of the secret, never the whole thing in the open
- what rotating it involves, and what else to check
- 04
coordinate
until it is fixedwe stay available, re-verify once a fix ships, and follow each organisation's own disclosure or bounty process where one exists.
- the cert or owner leads on timelines
- we re-download the new build and confirm the secret is gone
- nothing is ever published while the flaw is live
four rules.
report privately first
issues go to the owner, or to the national cert for public bodies. nothing goes public while a flaw is live.
static, read-only, good faith
we never exploit findings, access user data, or disrupt services. we confirm exposure; we never overstate it.
secrets stay secret
we never publish credentials or keys. full evidence is shared with the owner over a secure channel, under nda where asked.
coordinate the fix
we stay available, re-verify once a fix ships, and let the cert or owner lead on timelines.
what we never do.
these are the lines we do not cross, on any app, for any reason.
- exploit a finding, or go beyond what is needed to confirm exposure
- access, download or retain real user data
- disrupt, degrade or load-test a live service
- name an affected app, organisation or developer while a flaw is live
- publish a credential, key or anything that could be used to exploit an issue
- overstate impact: we state confirmed exposure and leave the severity call to the owner
- ask for payment to withhold or delay a report