GITHUB FLIPS COPILOT TRAINING TO OPT-OUT ON APRIL 24; COPILOT CLI 1.0.13 BRINGS MCP INFERENCE APPROVALS, REWIND, AND SPEEDUPS
GitHub will start training Copilot on user interaction data by default on April 24 while Copilot CLI ships notable agent/MCP improvements. GitHub plans to use ...
GitHub will start training Copilot on user interaction data by default on April 24 while Copilot CLI ships notable agent/MCP improvements.
GitHub plans to use Copilot interaction data—inputs, accepted outputs, code context, file names, and chat—for training by default starting April 24, with opt-out available; Business/Enterprise, students, and teachers are exempt per this ITdaily summary, while The New Stack says GitHub will also share data with Microsoft story.
On the tooling side, Copilot CLI 1.0.13 pre-releases add MCP server-initiated LLM inference gated by a user review prompt, fix BYOM reasoning-effort handling, harden OAuth/allowlist flows, improve grep memory behavior, drop gemini-3-pro-preview, introduce a conversation timeline picker, make MCP registry lookups more reliable, and speed startup via a V8 compile cache (1.0.13-0, 1.0.13-1).
Default-on training changes data governance and IP risk for any developer using Copilot outside Enterprise/Business.
CLI updates make agentic workflows more practical and auditable with MCP approval prompts, rewind, and performance fixes.
-
terminal
Run a policy drill: identify any engineers on Free/Pro/Pro+ and validate opt-out settings; confirm Enterprise/Business exemptions cover your org use cases.
-
terminal
Upgrade to copilot-cli 1.0.13 pre-release in a sandbox and test MCP inference approval flow, BYOM reasoning-effort behavior, and grep with large files.
Legacy codebase integration strategies...
- 01.
Harden guardrails: enforce SSO and block personal GitHub accounts on corp repos; document Copilot usage and opt-out expectations.
- 02.
Pin or validate model choices after gemini-3-pro-preview removal, and verify OAuth/allowlist policies for MCP servers still behave as expected.
Fresh architecture paradigms...
- 01.
Adopt MCP with explicit approval prompts to keep agent actions observable; enable timeline rewind for debuggability.
- 02.
Start with a low-sensitivity repo and BYOM provider to shape prompts, logging, and review workflows before wider rollout.