Across this year's penetration tests of web apps, APIs and ERP systems, a few patterns showed up again and again.
- Broken object-level authorisation in APIs (IDOR)
- Over-privileged service accounts and leaked API keys
- Prompt injection in newly added AI assistants
- Missing rate limits on authentication endpoints
- Outdated third-party modules with known CVEs
What to do about it
Shift security left: threat-model new features, add SAST and dependency scanning to CI and schedule a manual test before every major release. Treat AI features as untrusted input processors, because that's what they are.



