Azure AI Content Safety / Prompt Shields: Pricing, Tiers and What's Actually Covered
Microsoft Azure native content-safety stack including Prompt Shields (jailbreak / user-prompt-attack detection), Groundedness, Task Adherence and harm-category filters.
Pricing tiers
Verbatim from the Azure Prompt Shields pricing page on 2026-06-19. Quote-only tiers are surfaced as such, never with an inferred price.
| Tier | Price | Free allowance | Notes |
|---|---|---|---|
| F0 free tier | Free / request | 5 RPS, capped monthly | Includes Prompt Shields |
| S0 standard tier | $0.38 per 1K text records | - | 1 record = up to 1,000 unicode chars; Prompt Shields, Groundedness, Protected Material, Task Adherence |
Source: https://azure.microsoft.com/en-us/pricing/details/cognitive-services/content-safety/ | Verified 2026-06-19
OWASP LLM Top 10 coverage
What Azure Prompt Shields defends against, as claimed on its product pages and security documentation. Coverage flags are binary: an item is listed only if the vendor names the attack class explicitly.
- LLM01 Prompt injection
- LLM02 Sensitive information disclosure
- LLM03 Supply chain
- LLM04 Data and model poisoning
- LLM05 Improper output handling
- LLM06 Excessive agency
- LLM07 System prompt leakage
- LLM08 Vector and embedding weaknesses
- LLM09 Misinformation
- LLM10 Unbounded consumption
Hidden costs
Beyond the line-item price, every runtime AI guardrail carries operating costs the vendor page does not surface. Plan for these in any board paper.
- Per-call latency overhead. Guardrail check sits in the request path; budget 50 to 300 ms extra per LLM call.
- False-positive remediation. Allocate engineering time to tune policies after every guardrail rule change.
- Logging and retention. Guardrail decisions need to land in your SIEM. See siemcostcalculator.com.
- On-call coverage. Treat AI security alerts like SOC alerts. See securityoperationscost.com.
What Azure Prompt Shields is best for
Microsoft Azure native content-safety stack including Prompt Shields (jailbreak / user-prompt-attack detection), Groundedness, Task Adherence and harm-category filters.