Microsoft Disables Copilot Vision in Edge After Privacy Backlash
Microsoft abruptly halted the rollout of Copilot Vision in Microsoft Edge on November 28, 2025, disabling the feature that allowed the AI assistant to view and narrate users’ browser tabs in real time. The decision followed widespread criticism that the opt-in setting was buried deep in preview builds and that recorded sessions were retained on Microsoft servers longer than advertised. Internal telemetry revealed that fewer than 3 percent of Insiders had actively enabled the capability, yet over 180,000 screen-capture sessions had already been uploaded for “improvement purposes” before the shutdown.
Copilot Vision, introduced at Ignite 2025 as a multimodal extension of Copilot for Microsoft 365, combined GPT-4o vision capabilities with live browser context to answer questions like “summarize this 10-K filing” or “find the cheapest flight on this page.” The implementation relied on continuous frame sampling every 300 milliseconds, encrypted in transit but stored unencrypted in Azure blobs for up to 30 days under the default diagnostics policy. Privacy advocates flagged that the feature captured sensitive data—including banking portals and medical records—without explicit per-site consent mechanisms required by GDPR and California CCPA regulations.
Microsoft confirmed the immediate suspension applies globally across Edge Stable, Beta, and Dev channels, with all previously collected visual data scheduled for deletion by December 5, 2025. A company spokesperson stated the pause allows redesign of consent flows and introduction of on-device processing for future iterations. Build 131.0.2924.0 and later automatically disable the Copilot Vision toggle and purge local caches. Enterprise tenants enrolled in Microsoft 365 Copilot remain unaffected, as the feature never reached production for commercial SKUs.
The incident triggered an expedited review by the Irish Data Protection Commission, the lead supervisory authority for Microsoft in the EU, focusing on lawful basis for processing under Article 6. Regulators requested audit logs showing that 11 percent of sampled sessions contained personally identifiable information despite Microsoft’s earlier claims of automatic redaction. In parallel, the U.S. Federal Trade Commission opened an inquiry into whether the diagnostics labeling accurately reflected the scope of data collection.
Developers report that the underlying vision pipeline has been disconnected from the Edge sidebar, with residual code gated behind a hidden feature flag set to false. Microsoft published an updated privacy addendum acknowledging that “certain preview experiences may collect broader telemetry than production features” and committed to client-side inference using Phi-4 Vision models for any resumption. Industry observers note the setback delays Microsoft’s answer to Google’s Gemini Live screen-sharing and OpenAI’s planned Canvas browser integration, both of which launched with stricter opt-in boundaries.
As of November 29, 2025, Copilot in Edge reverts to text-only context from tabs, preserving URL and title metadata but no longer accessing rendered pixels. Microsoft plans to reintroduce a revised version no earlier than February 2026, contingent on independent privacy assessment and explicit per-domain permissions. The episode underscores mounting regulatory pressure on real-time multimodal assistants operating inside browsers, where traditional sandbox protections do not extend to AI-driven observation.
