Varselio vs Lakera, SafePrompt and Azure Prompt Shields
Varselio is a Swedish security service that checks what your website chatbot or AI agent is about to say or do, in real time, and stops prompt injection, jailbreaks and data leaks before they cause harm, asking a human before risky actions.
All four products help protect AI chatbots and agents against prompt injection. They are built for different buyers. Below, the other products are described only with facts from their own public pages, checked on 25 October 2026. Prices and features change, so check each vendor's page before you decide.
| Varselio | Lakera | SafePrompt | Azure Prompt Shields | |
|---|---|---|---|---|
| What it is | Real-time checks of a chatbot's or agent's messages and actions, with human approval for risky ones, traps and an evidence log | “AI Agent Security”: agent discovery, risk assessment and runtime protection | A prompt injection detection API (one POST call per text) | A detector for user prompt attacks and document attacks in Azure AI Content Safety and Microsoft Foundry |
| Built for | Companies of any size, including small businesses without a security team | Enterprise security teams | Developers of LLM apps and agents | Teams building on Azure |
| Price (public) | Free start; prepaid passes 279 / 1,419 / 2,259 SEK incl. VAT for 1 / 6 / 12 months; nothing renews | Not listed on the product page | Free ($0, 10,000 validations/month), Starter $29/month, Business $99/month; paid plans renew monthly | Not listed on the Prompt Shields documentation page (see Azure pricing) |
| Blocks or only reports? | Answers go ahead, wait for a human, or stop | Detects and blocks prompt attacks, data leakage and unsafe agent actions | Returns a verdict; your app enforces it | The Content Safety API returns results and your app decides; in Foundry, blocking can be turned on |
| Human approval step | Yes, on your phone | Not described on the page | Not described on the page checked | Not described on the page |
| Languages | English and Swedish | Not stated on the page | Not stated on the pages | Primarily trained and evaluated on English content |
| Where it is based | Sweden; data hosted in the EU | Lakera Inc, San Francisco | Not stated on the pages checked | Microsoft Azure |
When to choose which
- Varselio if you want a ready service that both checks and acts (stop, ask a human, allow), a dashboard and alerts, simple prepaid pricing in SEK, EU data hosting and a Swedish-speaking contact. Varselio is new and in public beta, so it has a shorter track record than the others.
- Lakera if you are an enterprise security team that needs to discover and govern many AI agents across the organisation.
- SafePrompt if you are a developer who wants a detection API with a free monthly quota and will build the enforcement yourself.
- Azure Prompt Shields if your AI already runs on Azure or Microsoft Foundry.
Sources: lakera.ai/lakera-guard, safeprompt.dev/pricing, Microsoft Learn: Prompt Shields. Spotted an error? E-mail dennis@varselio.se and it will be corrected.
About prompt injection (OWASP LLM01)
Prompt injection is number one on the OWASP Top 10 for LLM applications: LLM01:2025 Prompt Injection. OWASP notes that it is unclear whether fool-proof prevention exists, and recommends layered measures such as input and output filtering, least privilege and human approval for high-risk actions.
Varselio reduces risk, but no security product is 100% secure: it cannot guarantee that every harmful message or action is detected.