Guide EN-05
AI API Relay Risks
AI API relays can be useful, but users should check pricing anomalies, model substitution, prompt logging, balance loss, and enterprise compliance boundaries.
Reasonable use
Low-risk, repeatable, non-sensitive personal tasks.
Verify first
Model behavior, latency, error rate, refunds, terms, privacy, and ownership.
Avoid
Customer data, company source code, internal docs, and production decisions.
Risk checklist
These signals matter more than the discount.
| Risk | Signal | How to verify | Impact | Action |
|---|---|---|---|---|
| Model substitution | Claimed model underperforms | Compare fixed prompts with official API | Unreliable output | High risk |
| Balance loss | No refunds or unreachable support | Check history and refund rules | Financial loss | High risk |
| Prompt logging | No retention policy | Read privacy and logging terms | Data exposure | High risk |
| Extreme pricing | Far below rational cost | Compare official and aggregator prices | Unknown supply | Needs explanation |
Self-check
Spend ten minutes before adding balance.
- Check legal entity, terms, privacy policy, refund rules, and update date.
- Run fixed prompts with small requests before larger tasks.
- Test streaming, rate limits, and error behavior.
- Do not upload private source code or customer data.
Use boundary
Personal low-risk
Small tests can be reasonable.
Team development
Prefer BYOK or official channels.
Enterprise buying
Require contracts, DPA, SLA, and invoices.
Public ranking
Show evidence and verification dates.
Submit risk sample