Spaces:
Sleeping
Sleeping
Create governance-site/wiki/hf-trust-safety-justification.md
Browse files
governance-site/wiki/hf-trust-safety-justification.md
CHANGED
|
@@ -1,76 +1,46 @@
|
|
| 1 |
-
# Hugging Face Trust & Safety Justification
|
| 2 |
-
**Federal FOIA Intelligence Search (FFIS)**
|
| 3 |
|
| 4 |
-
|
| 5 |
-
|
| 6 |
-
|
| 7 |
-
|
| 8 |
-
|
| 9 |
-
|
| 10 |
-
AI usage is:
|
| 11 |
-
- Optional
|
| 12 |
-
- User-initiated
|
| 13 |
-
- Formatting-only
|
| 14 |
-
|
| 15 |
-
---
|
| 16 |
-
|
| 17 |
-
## Safety Scope
|
| 18 |
-
|
| 19 |
-
FFIS:
|
| 20 |
-
- Uses public FOIA reading rooms only
|
| 21 |
-
- Does not scrape or ingest private data
|
| 22 |
-
- Does not profile individuals
|
| 23 |
-
- Does not generate intelligence assessments
|
| 24 |
|
| 25 |
---
|
| 26 |
|
| 27 |
-
##
|
| 28 |
-
|
| 29 |
-
-
|
| 30 |
-
|
| 31 |
-
- Predictive analytics
|
| 32 |
-
- Law enforcement targeting
|
| 33 |
-
- Facial recognition
|
| 34 |
-
- Behavioral inference
|
| 35 |
|
| 36 |
---
|
| 37 |
|
| 38 |
-
##
|
| 39 |
-
|
| 40 |
-
-
|
| 41 |
-
-
|
| 42 |
-
-
|
| 43 |
-
- No third-party data brokers
|
| 44 |
|
| 45 |
---
|
| 46 |
|
| 47 |
-
##
|
| 48 |
-
|
| 49 |
-
-
|
| 50 |
-
-
|
| 51 |
-
-
|
| 52 |
-
- AI disclosure footers
|
| 53 |
|
| 54 |
---
|
| 55 |
|
| 56 |
-
##
|
| 57 |
-
|
| 58 |
-
|
| 59 |
-
|
| 60 |
-
| Misuse | Hard governance flags |
|
| 61 |
-
| Over-reliance | Disclaimers |
|
| 62 |
-
| Privacy | Public-only sources |
|
| 63 |
-
| Legal misuse | No filing integration |
|
| 64 |
|
| 65 |
---
|
| 66 |
|
| 67 |
## Conclusion
|
| 68 |
-
|
| 69 |
-
|
| 70 |
-
-
|
| 71 |
-
|
| 72 |
-
-
|
| 73 |
-
- Governance
|
| 74 |
-
- Responsible AI use
|
| 75 |
-
|
| 76 |
-
---
|
|
|
|
| 1 |
+
# Hugging Face Trust & Safety Justification
|
|
|
|
| 2 |
|
| 3 |
+
## System Classification
|
| 4 |
+
Public-record aggregation tool
|
| 5 |
+
No private data ingestion
|
| 6 |
+
No model training
|
| 7 |
+
No surveillance capability
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 8 |
|
| 9 |
---
|
| 10 |
|
| 11 |
+
## Content Sources
|
| 12 |
+
✔ Public FOIA electronic reading rooms
|
| 13 |
+
✔ Government-owned domains only
|
| 14 |
+
✔ No scraping behind authentication
|
|
|
|
|
|
|
|
|
|
|
|
|
| 15 |
|
| 16 |
---
|
| 17 |
|
| 18 |
+
## AI Risk Mitigation
|
| 19 |
+
- AI disabled by default where feasible
|
| 20 |
+
- Formatting-only AI usage
|
| 21 |
+
- Explicit opt-in controls
|
| 22 |
+
- Hard feature flags enforced in code
|
|
|
|
| 23 |
|
| 24 |
---
|
| 25 |
|
| 26 |
+
## Abuse Prevention
|
| 27 |
+
- No bulk downloading automation
|
| 28 |
+
- No scraping circumvention
|
| 29 |
+
- No identity inference
|
| 30 |
+
- No profiling or targeting
|
|
|
|
| 31 |
|
| 32 |
---
|
| 33 |
|
| 34 |
+
## Transparency Measures
|
| 35 |
+
- Cryptographic provenance headers
|
| 36 |
+
- Public governance documentation
|
| 37 |
+
- Explicit disclaimers in UI and PDFs
|
|
|
|
|
|
|
|
|
|
|
|
|
| 38 |
|
| 39 |
---
|
| 40 |
|
| 41 |
## Conclusion
|
| 42 |
+
This system aligns with HF Trust & Safety policies for:
|
| 43 |
+
✔ Transparency
|
| 44 |
+
✔ Non-harmful use
|
| 45 |
+
✔ Public data only
|
| 46 |
+
✔ Human-initiated actions
|
|
|
|
|
|
|
|
|
|
|
|