Lasso: 20,000+ once-public GitHub repositories from Microsoft, AWS, Google, IBM, Tencent, and 16,000+ other companies are still accessible via Copilot
Security researchers are warning that data exposed to the internet, even for a moment, can linger in online generative AI chatbots …
Context & Ripple Effects
This report extends a recurring GitHub exposure problem from raw repository access into AI-mediated retrieval. Earlier coverage of a Microsoft AI unit repository exposing 38TB of sensitive material showed the immediate risk of accidentally public code and credentials; the new finding suggests that changing a repository’s visibility may not end every downstream access path.
The affected set spans major cloud and technology companies as well as thousands of other organizations, making repository lifecycle controls—not only initial publication status—a material concern for teams using Copilot.
First-order effects
- Organizations whose repositories were once public face a continuing risk that code or data they expected to be withdrawn can be surfaced through Copilot.
- Microsoft and GitHub must address a mismatch between repository visibility changes and Copilot-accessible content, while affected companies may need to reassess what was exposed before repositories were made private or removed.
Second-order effects
- Security and developer-platform teams are likely to treat AI coding assistants as an additional access layer in incident response, alongside source-control permissions, secret rotation, and repository takedowns.
- The finding raises the value of auditable deletion, indexing, and retention controls for AI-assisted developer tools, rather than relying on a repository status change alone.
Third-order effects
- If similar cases persist, software supply-chain security will increasingly define exposure by what AI systems can retrieve, not just by what remains publicly reachable on the web.
- The broader shift is toward governed AI corpora: providers may be pressed to make provenance, removal handling, and access boundaries more transparent for enterprise customers.
The trend: Generative AI is turning historical data exposure into a durable access-governance problem, requiring controls over model retrieval and retention as well as original publication.