Security researchers have uncovered a troubling privacy leak in Microsoft Copilot. Data exposed to the internet—even briefly—can persist in AI chatbots long after being made private.
Israeli cybersecurity firm Lasso discovered this vulnerability when their own private GitHub repository appeared in Copilot results. The repository had been accidentally public for a short time before being locked down.
"Anyone in the world could ask Copilot the right question and get this data," warned Lasso co-founder Ophir Dror.
The problem extends far beyond Lasso. Their investigation found over 20,000 since-private GitHub repositories still accessible through Copilot, affecting more than 16,000 organizations including Google, IBM, PayPal, Tencent, and Microsoft itself.
The exposed repositories contain damaging materials: confidential archives, intellectual property, and even access keys and tokens. In one case, Lasso retrieved contents from a deleted Microsoft repo that hosted a tool for creating "offensive and harmful" AI images.
Microsoft classified the issue as "low severity" when notified in November 2024, calling the caching behavior "acceptable." While the company stopped showing Bing cache links in search results by December, Lasso says the underlying problem persists—Copilot still accesses this hidden data.
Why this matters:
- The digital world has no true delete button. What you expose today might be repeated by AI tomorrow.
- Your "deleted" data isn't gone—it's cached in AI systems with perfect memory and questionable discretion.
- The traditional web security model is breaking down when AI can recall and share content that's no longer publicly available.
Read on, my dear:
