French AI startup Mistral AI released Large 4 to the public on 29 October, Channel News Asia reported. Mistral says the model is designed to help organisations identify and fix cybersecurity vulnerabilities. It can be downloaded and run on an organisation’s own systems, rather than accessed mainly through an online service, as Google’s Gemini is.
Checking a computer system for security weaknesses could therefore involve running the model on that system’s organisation’s own equipment, without relying on a third-party service for access. Mistral says that arrangement could reduce dependence on an outside provider.
Mistral says the public release improves on a version unveiled three weeks earlier, following tests with governments and corporate users. The company describes Large 4 as its most capable openly available model for cybersecurity use and says Chinese models dominate the field. Pierre Stock, Mistral’s vice-president of science, said he was uncomfortable with a world without open weights. He said that, once released, access to a model cannot be contained: users cannot be cut off, but anyone can use it.
