Leaked Presentation Reveals Microsoft's Early Work On Security Copilot, Potential And Pitfalls Of GPT-4

The creation of Microsoft Corporation’s MSFT Security Copilot, an AI-driven security service, illustrates the potential and challenges of generative AI, new information reveals.

What Happened: Based on an excerpt of a presentation by Microsoft from late 2023, Business Insider reported on Tuesday that the software producer rolled out its Security Copilot service in early 2023, utilizing OpenAI’s GPT-4 and an in-house model to tackle cybersecurity queries. The presentation disclosed the hurdles and breakthroughs involved in developing this pivotal AI product.

Lloyd Greenwald, a Microsoft Security Research partner, unveiled the initial focus on Microsoft’s internal machine-learning models for security tasks. However, due to the widespread use of Microsoft’s GPUs for GPT-3 work, the project was hampered by a shortage of computing resources.

Microsoft then gained early access to GPT-4, prompting a shift in focus from its models to GPT-4 in cybersecurity. The company showcased its early GPT-4 explorations to government and external customers, emphasizing the advantages of using a single universal AI model.

Microsoft continues to utilize specific machine-learning models for particular issues, such as attack campaign attribution and compromised account detection. The company’s spokesperson, Frank Shaw, stated that the meeting pertained to technology built on GPT 3.5 and is “irrelevant” to the current Security Copilot built on GPT-4.

Why It Matters: Microsoft’s journey to develop Security Copilot sheds light on the possibilities and challenges of generative AI, especially with GPT-4, which is renowned for its advanced natural language processing capabilities.

Microsoft unveiled its AI-driven security assistant in early 2023, with qualified users getting early access in October.

Image Credits – Shutterstock