Add content from: Hacking Google with AI: AI-Assisted Discovery-Document Fuzzi...

This commit is contained in:
HackTricks News Bot
2026-06-11 16:13:40 +00:00
parent b81257def6
commit 108413b913
@@ -75,10 +75,111 @@ Help Token: ARD_zUaNgNilGTg9oYUnMhfa3foMvL7qspRpBJ-YZog8RLbTjCTBolt_WjQQ3myTaOqu
reason: AUTH_PERMISSION_DENIED
```
### Brute Force API endspoints
### Discovery-document driven API recon
Once you have a valid key, don't just test the obvious public APIs. In Google environments, **discovery documents** can work as a machine-readable attack-surface map exposing **resources, methods, paths, HTTP verbs, parameters, and request/response schemas**.
Useful targets:
```bash
# Public / common path
curl -s 'https://serviceusage.googleapis.com/$discovery/rest?key=<api-key>'
# Hidden docs behind a visibility label
curl -s 'https://serviceusage.googleapis.com/$discovery/rest?labels=GOOGLE_INTERNAL&key=<api-key>'
```
Notes:
- Compare the document size and method count with and without labels such as `GOOGLE_INTERNAL`.
- A **true nonexistent method** usually returns an **HTML 404**. A **JSON 404** with `Method not found.` can indicate that the method exists but the **API key project is missing a required visibility label**.
- Some first-party APIs are easier to reach via `*.clients6.google.com` or undocumented `*-pa.googleapis.com` hosts than via the public API explorer.
This is specially useful to enumerate **internal/admin APIs accidentally exposed to the Internet** and to build custom fuzzers from the schema instead of guessing JSON fields manually.
### Key ownership filtering at scale
If you collect many Google API keys from APKs, browser traffic, IPAs or binaries, you may want to quickly identify **which project owns each key** and whether it belongs to the target company.
A practical trick is to intentionally call an API that is **not enabled** for the key. Google frequently leaks the backing **project number** in the error:
```bash
curl -s 'https://protos.googleapis.com/$discovery/rest?key=<api-key>'
```
Look for messages like:
```text
Protos API has not been used in project 244648151629 before or it is disabled
```
That project number can then be correlated with other endpoints, leaked metadata, or internal company-mapping logic to separate **in-scope first-party keys** from customer / third-party projects.
### Handle key restrictions before discarding a key
A restricted key is not necessarily useless. Preserve the context where the key was found and replay requests with the matching headers:
```bash
# Browser restricted
curl -H 'X-Goog-Api-Key: <api-key>' \
-H 'Referer: https://target.google.com' \
'https://servicemanagement.googleapis.com/v1/operations'
# iOS restricted
curl -H 'X-Goog-Api-Key: <api-key>' \
-H 'X-Ios-Bundle-Identifier: com.google.GoogleMobile' \
'https://servicemanagement.clients6.google.com/v1/operations'
# Android restricted
curl -H 'X-Goog-Api-Key: <api-key>' \
-H 'X-Android-Package: com.google.android.settings.intelligence' \
-H 'X-Android-Cert: <sha1-signing-cert>' \
'https://servicemanagement.clients6.google.com/v1/operations'
```
Common restriction classes:
- **HTTP Referer**
- **iOS bundle ID** (`X-Ios-Bundle-Identifier`)
- **Android package + signing cert** (`X-Android-Package` + `X-Android-Cert`)
- **Server IP** restrictions, which generally cannot be bypassed remotely
### First-party auth / origin whitelist hints
Some Google web APIs accept a mix of **session cookie + first-party authorization headers** and are hosted on `*.clients6.google.com`. When testing these APIs:
- Keep `Origin` and `Referer` compatible with the product that normally uses the API.
- A `401` with `reason: SESSION_COOKIE_INVALID` can actually mean the **origin is not whitelisted**, not that the cookie is bad.
- Publicly reachable APIs that only accept internal origins such as `*.corp.google.com` are high-value candidates for broken access control review.
### Brute force enabled APIs and fuzz documented methods
As you might not know which APIs are enabled in the project, it would be interesting to run the tool [https://github.com/ozguralp/gmapsapiscanner](https://github.com/ozguralp/gmapsapiscanner) and check **what you can access with the API key.**
After you identify reachable services, prefer **schema-driven fuzzing** over blind guessing:
- Generate requests directly from discovery docs
- Replay the same request with **all known working keys**
- Normalize errors so you can distinguish **invalid input** from **visibility-label / auth / restriction** failures
- For numeric object identifiers, try nearby values such as `ID-1`, `ID+1`, `1`, `2`, `100`, `1000`
- Only report when you confirm **cross-user data access** or another real authorization failure; plain enumeration is usually not enough by itself
### Common vuln pattern: unauthenticated direct-object reference
A recurring bug pattern is an **internal/admin API** that accepts a victim-controlled identifier (for example `gaiaId`) and only validates that the request carries a usable API key, but **never checks whether the caller is authorized to access that specific object**.
```bash
curl 'https://gfibervoice-pa.googleapis.com/v1/BssGetVoiceSettings?gaiaId=<victim_gaia_id>' \
-H 'X-Goog-Api-Key: <api-key>'
```
If the endpoint returns another user's phone number, email address, settings, or other private fields, treat it as a classic **IDOR / broken access control** issue.
## References
- [Hacking Google with AI: AI-Assisted Discovery-Document Fuzzing of Google APIs](https://brutecat.com/articles/hacking-google-with-ai)
- [Google APIs Explorer](https://developers.google.com/apis-explorer)
{{#include ../../../banners/hacktricks-training.md}}