Get the app →
BriefTea logoBriefTea

Every story in sixty words

Get the free app no ads · no paywall

Explained in plain English

What does a cyber test measure?

When Mistral says some closed models scored near zero on a cyber test, it means they were highly vulnerable to malicious attacks or manipulation. These tests assess an AI's resilience against attempts to extract sensitive data, inject harmful instructions, or make the model behave unexpectedly. A low score indicates significant security weaknesses.

Stories that explain this

Mistral launches new AI model, pitching open weights for security

Related explainers

Why would AI firms share their code? Why is France a big AI player? How can a vendor limit an AI model? How do AI companies test? What are AI 'test environments'? How do AIs attempt cyber-attacks? How do apps test new features? Your bank's cyber defences

The full catalogue

Browse every card filed under C →