Radar #12
Sep 7–13, 2026 · Published Sep 7, 2026, 5:40 PM UTC
Fleet snapshot
Anonymous counts across all complete scans this week. No repository names are shown.
498 repositories were scanned for AI-context hygiene.
1727 AI-context risk signals matched 19 sensitive path categories.
1% of scanned repos use a dedicated AI exclude file (3 types in use).
How to read this report
These are not confirmed leaks.
Radar uses path-level checks to detect files and folders that may become visible to AI coding tools if they are not covered by AI-specific ignore rules.
A risk signal is a file committed to the repository whose path matches a sensitive category (for example .env or *.pem). Counts are file matches at the path level — Radar never opens the files.
- No file contents were read.
- No secrets were collected.
- No repository names are published without opt-in.
- Repositories are a manually curated cohort, not a random sample — read these as directional signals, not population-level statistics.
Scanned
AI-context risk signals
Repos with risk signals
AI exclude-file coverage
Scan details
- Targeted
- 500
- Scanned
- 498
- Scan complete
- 498
- Skipped
- 0
- Errors
- 2
- Repos with risk signals
- 128
- Generated
- Sep 7, 2026, 5:40 PM UTC
- Tool version
- 0.21.0–0.21.0
- Ruleset
- e634955c141dcfc6
Main takeaway
26% of scanned repos had AI-context risk signals — mostly secret and signing path categories.
Far fewer repos use a dedicated AI exclude file (1%) than show risk signals (26%). Start by excluding Environment files (.env) and Key material (.pem, .key) from AI context.
AI-context risk signal categories
Aggregated path pattern counts — not tied to any named repository.
- Key material (.pem, .key)Required
- Key FilesRequired
- Environment files (.env)Required
- Kubernetes configRequired
- Secrets YamlRequired
- Pkcs12 P12Recommended
- Pkcs12 PfxRecommended
- npm credentials (.npmrc)Recommended
- Yarn ConfigRecommended
- Android KeystoreRecommended
- Auth JsonRecommended
- Firebase Client ConfigRecommended
- Ssh FilesRecommended
- Terraform VarsRecommended
- Pgp KeysRecommended
- Log FilesInfo
- Data ExportsInfo
- Backup FilesInfo
- Local DatabasesInfo
AI exclude files
1% of repos use a dedicated AI exclude file
These actually keep matching files out of AI context — and they are still rare.
Show 9 more exclude file types (3 in use overall)
AI context files
29% of repos ship an AI context file
Files like AGENTS.md and CLAUDE.md give AI tools project instructions — theyadd context, they do not exclude anything. They are not a substitute for an exclude file.
Fix the top risks
Patterns covering this week's most widespread categories (Environment files (.env), Certificates and PEM keys (.pem), npm credentials (.npmrc), Private keys (.key)). Add them to a dedicated AI exclude file (for example .cursorignore or.claudeignore) — and to .gitignore so the files are never committed.
.env
.env.*
!.env.example
*.pem
*.key
certs/
.npmrc
secrets/Per-category guidance lives in the risk pattern guides.
Compared with previous report
| Metric | Radar #11 | Radar #12 |
|---|---|---|
| Repositories scanned | 498 | 498 |
| AI exclude-file coverage | 1% | 1% |
| Repos with risk signals | 27% | 26% |
| AI-context risk signals (absolute) | 1727 | 1727 |
| Risk signals per scanned repo | 3.5 | 3.5 |
| Top category | Key material (.pem, .key) | Key material (.pem, .key) |
Named participants
Per-repository reports for maintainers who opted in. Fleet counts above stay anonymous; each card below is that repo's own scan.
No named listings this week. Maintainers can opt in via radar registry.
Check your own project
128 of 498 repos this week exposed sensitive paths to AI tools. Check yours in seconds:
Run the same local check with Offsend CLI:
brew install --cask offsend/tap/offsend-cli
offsend show
offsend prepare --dry-runEverything runs locally. No file contents are uploaded.