Microsoft released Security Copilot today. It is a new tool that aims to “summarize” and “make sense” of threat intelligence. This is part of Microsoft’s ongoing effort to add generative AI to all of its products.
In an announcement with few details, Microsoft pitched Security Copilot as a way to connect data about attacks and put security incidents in order of importance. This is already done by a lot of tools.
But Microsoft says that the generative AI models from OpenAI, especially the recently released text-generating GPT-4, make Security Copilot, which works with its other security products, better.
In a canned statement, Microsoft Security executive vice president Charlie Bell said, “To improve the state of security, you need both people and technology: human creativity and the most advanced tools that help people use their expertise quickly and on a large scale.” “With Security Copilot, we’re working toward a future where every defender has the tools and technologies they need to make the world safer.”
Strangely, Microsoft didn’t say exactly how GPT-4 is used in Security Copilot. It instead pointed out that Security Copilot is powered by a trained custom model, possibly based on GPT-4, that “incorporates a growing set of security-specific skills” and “deploys skills and queries” that are related to cybersecurity.
Microsoft emphasized that the model isn’t trained on customer data, which is a common complaint about services that use language models.
Microsoft says that this custom model helps “catch what other approaches might miss” by answering questions about security, recommending the best course of action, and summarizing events and processes.
But since text-generating models tend to lie, it’s hard to say how well they would work in production.
Microsoft admits that the custom Security Copilot model doesn’t always get everything right. “Mistakes can be made in content made by AI,” the company writes. “As we keep learning from these interactions, we are making changes to its answers to make them more coherent, useful, and relevant.”