Welcome to Alethia AI

​

What is Alethia AI?

Alethia AI helps you answer one important question:

“Does this AI model respond safely when people try to misuse it?”

You bring an LLM (like GPT-4o, Claude, Gemini, or your own custom model). Alethia sends it a library of carefully designed prompts that try to push it toward unsafe answers. Then a panel of independent AI judges grades each response. Finally, you get clear results, dashboards, and audit-ready reports. If your organization has to comply with the EU AI Act, Alethia also gives you the full risk management workflow you need — classification, risk register, mitigations, and exportable compliance reports.

​

Who is Alethia for?

Alethia is built for teams who need to trust the AI they ship — and prove that trust to regulators, customers, and leadership.

​

What can I do with Alethia?

• Test any LLM — OpenAI, Anthropic, Google, Mistral, DeepSeek, or your own custom endpoint.
• Use 137 ready-made prompts across 18 safety categories — or upload your own.
• Get verdicts you can trust — three independent AI judges grade every response.
• Override anything — humans always have the final say (and it’s all logged).
• Schedule recurring tests — catch regressions automatically.
• Compare models side by side — “Is GPT-4o safer than Claude for our use case?”
• Manage EU AI Act risks — classification, register, assessments, mitigations, and reports.

​

What this guide covers

This documentation is written for the person using Alethia, not the developer running the codebase. No coding required. If you can navigate a website, you can use Alethia. Use the sidebar on the left to jump to any topic, or follow the Quickstart to get hands-on right away.