EduTrainer security audit: developer finds hidden flaws in service code
A developer audited the EduTrainer educational service, uncovering admin panel vulnerabilities, broken deletion mechanisms, and undocumented AI data sharing.

In early November, a developer behind the educational platform EduTrainer conducted an internal code audit to uncover hidden architectural and logical issues. Even though the platform was running smoothly, tests were passing, and health checks returned positive results, inspecting the codebase revealed significant discrepancies between interface promises and backend behavior.
- The audit exposed security flaws in the staff administration panel, incomplete user data deletion, and unannounced data sharing with AI providers.
- Two-factor authentication (TOTP) secrets for staff members were stored in the database in plaintext without encryption.
- The developer patched these architectural flaws, updated the privacy policy, and added comprehensive test coverage.
Staff Panel Security Flaws and Plaintext 2FA Secrets Middleware inspection revealed a vulnerability in how staff routes were accessed. Requests starting with the /staff/ prefix bypassed the second-factor authentication (TOTP) check, meaning a compromised employee password was enough to gain full access. Furthermore, an investigation into the StaffTOTP model showed that 2FA secret keys were stored in plaintext directly in the database rather than being encrypted.
Incomplete Data Deletion and Privacy Policy Discrepancies While the account deletion feature removed profile data from the interface, it left complete texts of students' essays, exam attempts, and other educational artifacts lingering in the database. Additionally, analytics scripts tracked users before collecting proper consent, and the automated essay-checking feature sent full student texts to third-party AI models (OpenAI, DeepSeek, GigaChat) despite privacy documentation promising otherwise.
Technical Fixes and Lessons Learned for Developers Beyond logic errors, infrastructural vulnerabilities were addressed, including fixing a CSV import mechanism that loaded entire files into memory without size limits, and adjusting rate-limiting fallback behaviors during Redis outages. This audit highlights the importance of ensuring that a system's actual code matches its stated security and privacy guarantees.
Source: Все публикации подряд на Хабре



