JetBrains wants to train AI models on your code snippets (www.theregister.com)

JetBrains is asking organizations to share detailed code-related telemetry — code snippets, prompts and AI responses, edit histories and terminal usage — to help train its AI models, and is dangling one-year All Products Pack subscriptions (about $979/user/year) as an incentive. The new data-sharing option will appear in the 2025.2.4 releases of IntelliJ IDEA, PyCharm, Rider, RubyMine and PhpStorm (about two weeks away). For commercial-license users it’s opt-in; it will be on by default for some non-commercial users and off for centrally managed org licenses. JetBrains says public code corpora miss “complex, real-world” workflows and that a promising internal trial justifies scaling to external data. This matters because access to real development telemetry could materially improve code-completion, repair and context-aware assistants — but it raises concrete IP and privacy risks (including accidental leakage and model regurgitation of proprietary code, as seen in prior incidents). JetBrains already offers its own agent Junie and supports a multi-agent ecosystem via Anthropic’s Agent SDK and Claude 4.5 Sonnet, so the move signals a push toward building or fine-tuning proprietary models rather than relying solely on third-party LLMs. Practical implications: organizations must weigh the value of free licenses and potential token costs (AI Pro includes only 10 $1 credits/month) against legal, security and billing unpredictability before enrolling.