ai-security
1.8 million apps were already scanned for secrets. Assume yours was one of them.
Security researchers have documented hardcoded secrets in Android apps for years as a slow, manual research exercise. Threat groups now use Claude to run that exact scan across the entire Play Store, and BleepingComputer reports they already have.
Anthropic disclosed this month that multiple threat groups, including financially motivated actors and state-sponsored espionage groups linked to Russia and China, tried to abuse Claude to extract secrets from 1.8 million Android apps, according to BleepingComputer's reporting on the company's threat intelligence findings. The scale is the news. The underlying vulnerability is not new at all.
A known problem, done by hand
Security researchers have been documenting hardcoded secrets in Android apps since at least 2022, when a Cybernews study of 30,000 apps found 55.94 percent leaking API keys and database links, identified through weeks of manual reverse engineering on "mediocre infrastructure." A follow-up study focused on apps advertising AI features found the exposure worse: 72 percent of analysed apps contained at least one hardcoded secret, averaging 5.1 per app, with 81 percent tied to Google Cloud projects and 285 unauthenticated Firebase databases exposing over 200 million files. Producing findings at that scale took specialist teams weeks.
What an LLM changes about the same scan
None of this required a new exploit. Decompiling an APK and grepping for API keys, Stripe secrets and open storage buckets is a mechanical, repetitive task, exactly the kind an LLM automates well. What used to be a research project bounded by a firm's staff-hours is now a script that runs continuously against the entire Play Store, re-checked every time a developer ships an update. The attacker no longer needs to choose which app to target; they can scan all of them and sort the results by what they find.
A secret shipped inside a released APK should now be treated as already public, not as protected by the odds nobody will bother looking.
What changes for anyone shipping a mobile app
Assume any API key, service credential, or cloud token compiled into a client-side build has already been extracted, whether or not your app is a specific target. That means rotating any secret found in a build artefact rather than waiting for a breach notice, moving credential issuance server-side wherever the architecture allows it, and auditing cloud IAM policies on the assumption that a leaked key will be tried. App store review does not catch this class of exposure, so the correct control sits in your own release pipeline: a mandatory secrets scan before every build ships, not an annual audit.
Sources
Written by the Devence Lab research team.